KR102508952B1 - Electronic apparatus for allocating task to multiple robots and operation method thereof - Google Patents

Abstract

A method of allocating tasks to a plurality of robots by an electronic apparatus of the present disclosure includes a step of obtaining first information and second information indicating suitability for behaviors of the plurality of robots in a predetermined situation; a step of confirming first task information for allocating tasks to the plurality of robots based on the output of a first network into which the first information is input, and confirming second task information for allocating tasks to the plurality of robots based on the output of a second network into which the second information is input; a step of determining final task information for allocating tasks to the plurality of robots based on the first and second task information; and a step of providing the final task information.

Description

Electronic device for allocating tasks to a plurality of robots and its operation method

The present disclosure relates to an electronic device for allocating tasks to a plurality of robots and an operating method thereof.

The technology of assigning tasks to robots is one of the key technologies with promising prospects in future battlefields in preparation for future battlefields in order to control currently reduced human resources, increased contribution of robots, and multiple robots. The placement of manpower in the right place in the future battlefield is an important factor that determines victory or defeat on the battlefield, and it can contribute to the smoothness of human resources through the minimum manpower placement to achieve the purpose. Currently, the selection of multi-robot mission assignments is determined by reflecting human intuition and command doctrine based on collected data. Selection due to the final decision of a person can be accompanied by many differences such as human error, emotional difference, and skill level difference. As the convolutional deep neural network contributes greatly to improving mission assignment accuracy, we intend to apply it to multi-robot mission assignment and decision-making.

The disclosed embodiments are intended to disclose an electronic device and an operating method thereof for assigning tasks to a plurality of robots. The technical problem to be achieved by the present embodiment is not limited to the technical problems described above, and other technical problems can be inferred from the following embodiments.

A method of allocating tasks to a plurality of robots by an electronic device according to an embodiment, comprising: acquiring first information and second information indicating suitability for a behavior of a plurality of robots in a predetermined situation; Based on the output of the first network into which the first information is input, first task information for allocating tasks to the plurality of robots is checked, and tasks are assigned to the plurality of robots based on the output of the second network into which the second information is input. checking second task information for allocating; determining final task information for allocating tasks to a plurality of robots based on the first task information and the second task information; and providing final mission information.

In a method for assigning tasks to a plurality of robots by an electronic device according to an embodiment, the first information may include information in the form of a 3D tensor, and the second information may include 2D image information. can

In the method of allocating a task to a plurality of robots by an electronic device according to an embodiment, the acquiring step includes resource information of each of the plurality of robots, information about battlefield conditions, information about a fitness function for each battlefield situation, and resource information. - Generating first information and second information based on the information about the matching condition of the robot behavior.

In the method for assigning tasks to a plurality of robots by an electronic device according to an embodiment, a first network includes a task for allocating tasks to the plurality of robots based on information indicating a degree of suitability for a robot's behavior in a battlefield situation. It is a genetic algorithm-based network that outputs information, and the second network learns to output mission information for allocating missions to a plurality of robots based on information representing the degree of fitness for robot behavior in a battlefield situation. may be a convolutional neural network.

In a method for assigning tasks to a plurality of robots by an electronic device according to an embodiment, the second information is information generated by processing first information, the second information is obtained as input information of a second network, and the first obtaining output information of a first network, into which information is input, as target information for input information of a second network; and learning a second network based on the input information and the target information.

In a method for assigning tasks to a plurality of robots by an electronic device according to an embodiment, second information is obtained as input information of a second network, and task information of a skilled person corresponding to the second information is input information of the second network. Obtaining as target information for ; and learning a second network based on the input information and the target information.

In the method for assigning a task to a plurality of robots by an electronic device according to an embodiment, the plurality of robots include a first robot, a second robot, a third robot, and a fourth robot, and the first task information is 1-1 task value for allocating a task to 1 robot, 1-2 task value for allocating a task to a second robot, 1-3 task value for allocating a task to a 3 robot, and 4 1-4 task values for allocating a task to the robot, and the second task information includes a 2-1 task value for allocating a task to the first robot, and a second task value for allocating a task to the second robot. It includes a -2 task value, a 2-3 task value for assigning a task to the third robot, and a 2-4 task value for assigning a task to the fourth robot, and the final task information is provided to the first robot. A first final task value for assigning a task, a second final task value for assigning a task to a second robot, a third final task value for assigning a task to a third robot, and a task assignment to a fourth robot In the step of determining the fourth final task value, the first final task value is determined from the 1-1 task value and the 2-1 task value, and the 1-2 task value and the 2-2 task value are included. A second final task value is determined from the values, a third final task value is determined from the values of tasks 1-3 and tasks 2-3, and a fourth task value is determined from tasks 1-4 and tasks 2-4. It may include determining a final task value.

In the method of allocating a task to a plurality of robots by an electronic device according to an embodiment, the checking may include: a first reliability, a 1-1 task value, The 1-2 task value, the 1-3 task value, and the 1-4 task value are checked, and based on the output of the second network to which the second information is input, the second reliability, the 2-1 task value, The step of determining the 2-2 task value, the 2-3 task value, and the 2-4 task value, wherein the determining step comprises: based on the first reliability and the second reliability, the 1-1 task value. A first final task value is determined from the value and the 2-1 task value, a second final task value is determined from the 1-2 task value and the 2-2 task value, and the 1-3 task value and the second task value are determined. determining a third final task value from the -3 task value, and determining a fourth final task value from the 1-4 task value and the 2-4 task value.

In a method for assigning tasks to a plurality of robots by an electronic device according to an embodiment, a first reliability is calculated based on Cronbach's Alpha Coefficient of an output of a first network, and a second reliability is a first reliability. 2 can be calculated based on the outputs of the softmax layer of the network.

The method of allocating tasks to a plurality of robots by an electronic device according to an embodiment may further include transmitting final task information to the plurality of robots through a communication device.

An electronic device for allocating tasks to a plurality of robots according to an embodiment, comprising: a memory in which at least one program is stored; And by executing at least one program, obtaining first information and second information representing the suitability of the robot's behavior in a predetermined situation, respectively, and a plurality of robots based on the output of the first network into which the first information is input. Check first task information for allocating a task to a robot, check second task information for allocating a task to a plurality of robots based on the output of the second network into which the second information is input, and determine the first task information and An electronic device including a processor that determines final task information for allocating tasks to a plurality of robots based on the second task information and provides the final task information may be provided.

A computer-readable non-transitory recording medium recording a program for executing a method of allocating tasks to a plurality of robots according to an embodiment in a computer, wherein the method measures the suitability of the actions of a plurality of robots in a predetermined situation, respectively. obtaining first information and second information indicating; Based on the output of the first network into which the first information is input, first task information for allocating tasks to the plurality of robots is checked, and tasks are assigned to the plurality of robots based on the output of the second network into which the second information is input. checking second task information for allocating; determining final task information for allocating tasks to a plurality of robots based on the first task information and the second task information; and providing final mission information. A non-temporary recording medium may be provided.

According to the present disclosure, an electronic device may assign tasks to a plurality of robots by using a neural network trained on the basis of information indicating suitability of a plurality of robots' behaviors and a network based on a genetic algorithm in a predetermined situation. In particular, the electronic device may allocate a task to each robot based on information representing a degree of suitability for a behavior of a plurality of robots in a predetermined situation in consideration of individual characteristics of each of the heterogeneous multi-robots. Since the electronic device simultaneously uses a genetic algorithm-based network and a neural network, the amount of data required to assign tasks to a plurality of robots can be effectively reduced compared to the case of using only the genetic algorithm-based network. In addition, since the electronic device determines a final task value from a task value of the first network and a task value of the second network based on the first reliability of the first network and the second reliability of the second network, the final task with higher reliability is determined. By calculating the values, tasks can be more accurately assigned to multiple robots. The electronic device may provide improved task assignment performance through a convergence method in consideration of the reliability of the output of the first network through the first information and the output of the second network through the second information.

1 shows an electronic device according to the present disclosure.
2 shows a flowchart of a method of assigning a task to a unit robot.
3 shows an embodiment of generating information representing the degree of suitability of a robot's behavior in a given situation.
4A shows an example of first information that is input information of a first network.
4B shows an example of second information that is input information of a second network.
5 shows a convolutional neural network as an example of the second network.
6 shows an example of entropy for the outputs of the second network.
7 illustrates an embodiment in which a processor determines a first final task value, a second final task value, a third final task value, and a fourth final task value.
8 shows an embodiment in which an electronic device operates.
9 illustrates a method of operating an electronic device according to an exemplary embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

The terms used in the embodiments have been selected from general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but they may vary depending on the intention or precedent of a person skilled in the art, the emergence of new technologies, and the like. In addition, in a specific case, there are also terms arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the corresponding description. Therefore, terms used in the present disclosure should be defined based on the meaning of the term and the general content of the present disclosure, not simply the name of the term.

When it is said that a certain part "includes" a certain component throughout the specification, it means that it may further include other components without excluding other components unless otherwise stated. In addition, terms such as "..unit" and "..module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software. there is.

The expression of "at least one of a, b, and c" described throughout the specification means 'a alone', 'b alone', 'c alone', 'a and b', 'a and c', 'b and c' ', or 'all of a, b, and c'.

Hereinafter, with reference to the accompanying drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily carry out the present disclosure. However, the present disclosure may be embodied in many different forms and is not limited to the embodiments described herein.

In the present disclosure, a rule-based method such as a Genetic Algorithm (GA) is applied to implement a task assignment system prior to deep learning learning to assist a task assignment decision system of multiple robots, and derive results. The present disclosure derives a model by combining the results obtained through a genetic algorithm with a deep learning-based learning method to assist the decision maker in making decisions, and when unlearned data is input, multiple robots are based on the pre-learned model. You want to assign a task to

1 shows an electronic device according to the present disclosure.

The electronic device 100 includes a processor 110 and a memory 120 . In the electronic device 100 shown in FIG. 1 , only elements related to the present embodiments are shown. Accordingly, it is apparent to those skilled in the art that the electronic device 100 may further include other general-purpose components in addition to the components shown in FIG. 1 .

The electronic device 100 may assign a task to the robot in a predetermined situation. In detail, the electronic device 100 may assign tasks to a plurality of robots based on information indicating suitability of the behaviors of the plurality of robots in a battlefield situation.

The processor 110 serves to control overall functions of the electronic device 100 . For example, the processor 110 generally controls the electronic device 100 by executing programs stored in the memory 120 of the electronic device 100 . The processor 110 may be implemented as a central processing unit (CPU), graphics processing unit (GPU), or application processor (AP) included in the electronic device 100, but is not limited thereto.

The memory 120 is hardware that stores various types of data processed in the electronic device 100, and the memory 120 may store data processed in the electronic device 100 and data to be processed. Also, the memory 120 may store applications and drivers to be driven by the electronic device 100 . The memory 120 may include random access memory (RAM) such as dynamic random access memory (DRAM) and static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and CD-ROM. ROM, Blu-ray or other optical disk storage, hard disk drive (HDD), solid state drive (SSD), or flash memory.

According to an embodiment, the processor 110 may obtain information representing a degree of suitability of a robot's behavior in a predetermined situation. Here, the degree of fitness may mean the degree to which a robot's behavior is suitable for a given situation when a specific robot performs a specific action in a given situation, or it is used to determine how appropriate a robot's behavior is for a given situation. information can be displayed. For example, if a robot makes a 'counterattack' as an action in the situation of 'enemy discovery', the robot's action can be evaluated according to criteria such as time required, attack power, and relative distance, and according to the evaluation result The suitability for the behavior of the robot may be expressed as a first value for required time, a second value for attack power, and a third value for relative distance. According to an embodiment, the processor 110 may obtain first information and second information each indicating a degree of fitness for the behavior of a plurality of robots in a predetermined situation. For example, each of the first information and the second information may indicate a degree of suitability for the behavior of the first robot in a predetermined situation, and each of the first information and the second information may represent the behavior of the second robot in a predetermined battlefield situation. fit can be shown. Here, the predetermined situation may be a situation in a mission for a specific purpose or a situation in a battlefield, but is not limited thereto. When the predetermined situation is a battlefield situation, the predetermined situation may include 'enemy discovery', 'obstacle discovery', and 'reconnaissance request', but is not limited thereto. In addition, the action of the robot may include 'reaction attack', 'detour movement', 'support request', 'move', 'route request', 'reconnaissance', and 'ignore', etc. The degree of fitness may be expressed or evaluated as a value according to criteria such as 'additional required time', 'reconnaissance area', 'risk level', 'relative distance', 'time required', and 'attack power', but is not limited thereto. According to an embodiment, the first information may include information in the form of a 3D tensor, and the second information may include image information. In other words, the first information may include information representing the degree of fitness for the behavior of each of the plurality of robots in the form of a 3D tensor in a predetermined situation, and the second information may include information representing the behavior of each of the plurality of robots in the predetermined situation. It may include information showing the fitness for the 2D image. According to an embodiment, the second information may be generated by processing from the first information. Specifically, the processor 110 may generate second information through a process of processing first information expressed in the form of a 3D tensor into a form of a 2D matrix. In other words, the processor 110 may generate second information by processing the first information in the form of a two-dimensional matrix and imaging it. The processor 110 generates first information and second information based on resource information of each of a plurality of robots, information about a predetermined situation, information about a fitness function for each predetermined situation, and information about a resource-robot behavior matching condition. information can be generated. A process of generating information representing the degree of suitability of a robot's behavior in a given situation will be described in detail with reference to FIGS. 3, 4a, and 4b.

According to an embodiment, the processor 110 may acquire the first information and the second information from the outside through a communication device (not shown) in the electronic device 100 . According to another embodiment, the processor 110 may obtain the first information and the second information from the memory 120 .

2 shows a flowchart of a method of assigning a task to a unit robot.

In step S210, the unit robot may check the primary mission. For example, a unit robot may see a task assigned to it called "maneuver". Specifically, the unit robot may check a task value corresponding to the task and perform the task corresponding to the task value. For example, when the unit robot checks a task value corresponding to a task of “activating”, the unit robot may perform a task of “activating”.

In step S220, the unit robot may check whether a specific event has occurred in the battlefield. Here, the situation may be a situation pre-registered in the unit robot, and may include, for example, an enemy attack, enemy reinforcements arrival, and an enemy retreat. The unit robot may perform the primary task as it is when a specific situation does not occur, and may proceed to step S230 when a specific situation occurs.

In step S230, the unit robot may grasp the nature of a specific situation. The unit robot can receive battlefield information through a communication device or grasp the nature of a specific situation through directly collected battlefield information. In addition, the unit robot may grasp the nature of a specific situation through the electronic device 100 of the present disclosure.

In step S240, the unit robot may check whether mission replanning is necessary or not according to the characteristics of the identified specific situation. A case in which mission replanning is required may refer to a case in which it is confirmed that a mission different from the primary mission previously performed by a unit robot has to be performed. If it is confirmed that there is no need for re-planning of the mission, the unit robot can continue to perform the primary mission that was previously performed. The unit robot may check whether mission replanning is necessary or not through the electronic device 100 of the present disclosure. If it is determined that mission replanning is necessary, the unit robot may proceed to step S250.

In step S250, the unit robot may check the (1+n)th task. Specifically, when it is determined that mission re-planning is necessary in step S240, the unit robot may check the (1+n)th mission corresponding to the final mission information determined by the electronic device 100 of the present disclosure. Here, the (1+n)th task may mean a task assigned right after the nth assigned task. The unit robot may check a mission value corresponding to a mission assigned to itself and perform a mission corresponding to the mission value. For example, if tasks such as "retreat (movement with destination change)" and "engagement" are assigned as the (1+n) order missions, the unit robot will be assigned "retreat (movement with destination change)", You can check the mission value corresponding to "engagement" and perform missions such as "retreat (movement with destination change)" and "engagement".

In step S260, the unit robot may check whether another specific situation has occurred on the battlefield after performing the (1+n)th task assigned to it. When a specific situation different from the previous one occurs on the battlefield, the unit robot performs step S230 again, and when a specific situation different from the previous one does not occur, the unit robot may determine that the mission has been achieved.

Since assignment of tasks to unit robots is relatively simple compared to multi-robots, it may be possible to assign tasks through real-time monitoring with a small number of operating personnel. However, since multiple robots are generally composed of heterogeneous multiples, the characteristics of each platform are different, and there may be limitations in performing real-time monitoring by a small number of operating personnel. Therefore, in order to effectively control multiple robots, it can be said that an automation technology capable of relieving the operator's burden is essential. These automation technologies may include autonomous technologies (eg, environment recognition, autonomous movement, etc.) necessary to execute matters determined as a response to contingencies in the failsafe function for each platform and mission assignment technology.

As described above, assignment of tasks that must be provided synchronously to multiple robots is a difficult decision for both skilled and unskilled users. This is because it is a method that relies on the commanding experience of an experienced person based on data such as the current situation, rapidly changing environment, and disturbance, so it is not constant and can change depending on people's emotions and situations. Therefore, the present disclosure intends to create an AI-based model to assist human decision-making in advance, and assist multi-robot task allocation decision systems when similar learned environments and events arrive.

3 shows an embodiment of generating information representing the degree of suitability of a robot's behavior in a given situation.

According to an embodiment, the processor 110 outputs task information for allocating tasks to a plurality of robots based on first information representing the degree of suitability of the actions of the plurality of robots in a predetermined situation in order to improve task assignment performance. A first network based on a Genetic Algorithm may be used, and task information for allocating tasks to a plurality of robots is output based on second information representing the degree of suitability of the actions of the plurality of robots in a predetermined situation. A second network learned to do so can be used.

A genetic algorithm is an optimization methodology that mimics biological evolution. The genetic algorithm may genetically represent a solution that is a target of the optimization algorithm. The genetic algorithm has the advantage that it can be used in any application if a fitness function for evaluating a genetically expressed solution can be designed. In addition, the genetic algorithm has the advantage of being free from the local minima problem during optimization and greatly reducing the optimization time through parallelization when a GPU server or the like is available. However, the genetic algorithm is difficult to apply in a continuous space rather than a discrete space, and the overall optimization time may be greatly affected by the complexity of the goodness-of-fit function. On the other hand, the CNN-based learning model has the advantage that the deviation of the derivation time is small when inputting information similar to the learned data and relatively fast inference is possible compared to the GA algorithm. In addition, the CNN-based learning model can use 2D RGB-based images as input data, so it has the advantage of being able to apply the widely used object recognition SOTA (State-of-the-art). However, the CNN-based learning model may have a disadvantage that the quality of output values may deteriorate when inputting unlearned data having relatively low similarity, but a separate data augmentation method may be employed to overcome this. In the present disclosure, the GA algorithm-based network and the multi-channel information of the neural network are integrated, and based on the reliability measurement of the data input through each channel, the optimal combination of each network is sought to improve the performance of multi-robot assignment. want to do

According to an embodiment, the processor 110 may perform resource information of each of a plurality of robots, information about a predetermined situation, information about a fitness function for each predetermined situation, and information about a resource-robot behavior matching condition based on , It is possible to generate information representing the degree of suitability for the behavior of a plurality of robots in a given situation. The generated information may be information in the form of a 3D tensor.

Referring to FIG. 3 , the processor 110 performs multiple operations in a battlefield situation based on predefined resource information of each of a plurality of robots, real-time resource information, predefined battlefield situation information, and real-time battlefield situation information. It is possible to generate information representing the suitability of the robot's behavior. The generated information may be information in the form of a 3D tensor. According to an embodiment, the predefined resource information may include information about the type and capability of mission equipment, and information about operating distance and communication distance, and real-time resource information about location, speed, and status of mission equipment. information may be included. Information on the predefined battlefield situation may include information on fitness functions for each battlefield situation and information on matching conditions for resource-robot behavior. information may be included. According to an embodiment, the processor 110 generates information on currently available resources based on predefined resource information and real-time resource information of each of a plurality of robots, and generates information about a predefined battlefield situation of each of a plurality of robots. The current battlefield situation and battlefield information may be generated based on the information and the real-time battlefield situation information. Subsequently, the processor 110 may generate information (Information Tensor) having a 3D tensor form through the generated information on currently available resources and information on the current battlefield situation and battlefield. As a result, the processor 110 comprehensively connotes predefined information and real-time information to generate information representing the degree of fitness for the behavior of a plurality of robots in a battlefield situation, so that the different characteristics of each of the multiple robots are effectively reflected in the generated information. It can be. A description of information having a 3D tensor form will be described in detail with reference to FIG. 4A below.

4A shows an example of first information that is input information of a first network.

According to an embodiment, the processor 110 includes resource information of each of a plurality of robots, information about a predetermined situation (Event), information about a fitness function for each predetermined situation, and information about matching conditions between resource-robot behaviors. Based on this, it is possible to generate information representing the degree of fitness for the behavior of a plurality of robots in a given situation, and the generated information can be identified as first information, which is input information of a first network. For example, the processor 110 may check information having a 3D tensor shape generated in FIG. 3 as first information.

Referring to FIG. 4A , the first information may have a 3D tensor shape. Referring to FIG. 4A , the x-axis of the first information represents a resource, the y-axis represents a robot's behavior, and the z-axis represents a category. Resources on the x-axis mean available resources (i.e., the number of robots), robot actions on the y-axis mean actions the robot can take when a specific situation occurs, and categories on the z-axis mean robot actions when a specific situation occurs. It means the criterion for judging the goodness of fit. For example, if the size of the tensor is (10, 5, 6), there are 10 resources (i.e., the number of robots), and there are 5 actions that the robot can take. Goodness of fit means that it can be expressed or evaluated as a value according to six criteria. According to an embodiment, an information tensor, which is first information, may be divided into at least one sector. For example, referring to Figure 4a, the information tensor can be divided into three sectors. Each sector represents the suitability of the robot's behavior when a specific situation occurs. Specifically, the lower left sector represents the suitability of the robot's behavior when an 'enemy is found' situation occurs, the middle sector represents the suitability of the robot's behavior when an 'obstacle is found' situation occurs, and the upper right sector represents the suitability of the robot's behavior It shows the suitability of the robot's behavior when a 'reconnaissance request' situation occurs. The number written in the information tensor means the degree of fitness when it is assumed that the robot performs a specific action when a situation corresponding to each sector occurs. For example, in the case of the lower left sector, when an 'enemy is found' situation occurs and the robot performs a response attack action, it indicates that the relative distance is 8, the attack power is 5, and the required time is 2. Similarly, if the robot performs detour movement in the lower left sector, it indicates that the relative distance is 8, the attack power is 0, and the required time is 4. The degree of fitness may be the same or different when the robot performs different actions despite being in the same sector (i.e., even though the situation that occurred is the same). For example, in the lower left sector (when an 'enemy found' situation occurs), even if the robot performs a response attack, a detour, or a request for support, the relative distance to the enemy does not change, so the relative distance on the y-axis is all set to 8. can be the same In addition, when the robot performs detour movement, it is implied that it does not attack during detour movement, so the attack power of the y-axis becomes 0, and there may be a difference from the attack power when the robot performs a response attack or support request action. can Finally, since the time required for the robot to perform an action of detour movement is relatively longer than the action of a response attack, the time required on the y-axis is 2 when the robot performs the action of a response attack, but In the case of performing the action of moving, the required time on the y-axis may increase to 4. In FIGS. 4A and 4B, the degree of fitness is expressed as a value according to at least one criterion, and at least one criterion may be set differently for each situation, but the values and criteria described in FIGS. 4A and 4B are only examples and are not limited thereto. .

4B shows an example of second information that is input information of a second network.

According to an embodiment, the processor 110 may generate second information through a process of processing the first information. For example, the processor 110 may generate the second information of FIG. 4B by processing the first information in the form of a 3D tensor of FIG. 4A into a form of a 2D matrix and processing it into an image. According to another embodiment, the processor 110 may perform resource information of each of a plurality of robots, information about a predetermined situation, information about a fitness function for each predetermined situation, and information about a resource-robot behavior matching condition based on , It is possible to generate information representing the degree of suitability for the behavior of a plurality of robots in a given situation, and the generated information can be identified as second information, which is input information of a second network.

Since the second information (ie, information matrix) of FIG. 4B is the same as the information tensor of FIG. 4A except for the x-axis, descriptions of overlapping contents with those of FIG. 4A are omitted.

According to an embodiment, the processor 110 uses a deep-learning CNN technique based on label data when the second information (eg, the information matrix of FIG. 4B) is provided as input information to improve task assignment performance. Models can be derived by learning or training. In addition, when new unlearned input information is provided to the processor 110, tasks may be assigned to a plurality of robots based on previously learned or trained models. As a method of allocating tasks through learning (based on CNN deep learning), a method of pairing input information with the correct answer is mainly used. In order to implement this, the processor 110 assigns the result obtained through the GA algorithm to the answer sheet to train the second network, or assigns the task result assigned by the expert in the operating environment to the answer sheet to train the second network to derive the model can do. According to an embodiment, the processor 110 may obtain second information representing a degree of suitability of a robot's behavior in a predetermined situation as information for learning or training a second network.

Referring to FIG. 4B , according to an embodiment, the processor 110 comprehensively connotes predefined information and real-time information like the first information to generate second information representing the suitability of the robot's behavior in a given situation. can do. According to another embodiment, the processor 110 may generate second information by processing and imaging previously generated first information. For example, if the size of the information tensor is (4,8,8), there are 4 resources (i.e., the number of robots) and there are 8 actions that the robot can perform. This means that the fitness for can be judged by eight criteria. Therefore, when converting the first information (i.e., the information tensor of FIG. 4a) into the second information (i.e., the information matrix of FIG. 4b), (4,8,8) becomes (8,8) * 4, and FIG. 4b Values corresponding to the y-axis and the z-axis may be obtained from information tensors of each of the first robot, the second robot, the third robot, and the fourth robot, like the information matrix of . The information matrix of FIG. 4B can be divided into at least one sector like the information tensor. Each sector may indicate the suitability of the robot's behavior when a specific situation occurs. For example, referring to Figure 4b, the information matrix can be divided into three sectors. The three sectors may be represented by RGB (Red, Green, and Blue). The processor 110 sets the number (minimum: 0, maximum: 10) written in the information matrix from the minimum value (0) to the maximum value of each sector color in order to use Backbone Architecture such as HR-Net32, linear-net, or ResNet50. It can be imaged by linear mapping with the value of (255). In addition, the processor 110 may generate second information by performing image resize to an image having a size of 224*224 pixels in order to use the information matrix as input information of the second network.

Referring to FIG. 4B , a table on the right side of the information matrix indicates tasks assigned to each of the first robot, the second robot, the third robot, and the fourth robot. Multi-robot mission assignment refers to an algorithm that derives countermeasures for various contingencies that occur while robots are performing their missions. The countermeasures derived in this way correspond to HLAC (High-level Action Command) in terms of robot control. For example, HLACs may include move, attack, dodge, or wait. The HLAC created in this way generates a Low-level Control Command (LLCC) to actually execute it. For example, if HLAC is given as 'move', the LLCC may generate and inform a route for performing 'move' using a gps value. As a countermeasure, HLAC can transform missions such as movement, attack, and avoidance into parameters of corresponding mission values. In addition, since the index of these variable task values can be expressed as a genetic expression required to apply the genetic algorithm, it is possible to derive an optimized countermeasure through the genetic algorithm. According to an embodiment, the processor 110 processes (ie, information matrix) the first information and obtains second information imaged as input information of the second network, and outputs the first network to which the first information is input. information (eg, the table on the right of FIG. 4B) may be acquired as target information for the input information of the second network, and the second network may be learned based on the input information of the second network and the target information of the second network. there is. In other words, the processor 110 indexes the output information of the first network and stores it as a json file, and trains the second network using the second information generated through the process of processing the stored json file and the first information can make it

Referring back to FIG. 1 , the processor 110 may check first task information for allocating tasks to a plurality of robots based on the output of the first network to which the first information is input, and the second information is input. Based on the output of the second network, second task information for allocating tasks to a plurality of robots may be checked. According to an embodiment, the first network may be a network that outputs task information for assigning tasks to a plurality of robots, and the second network may be trained to infer task information for assigning tasks to a plurality of robots. there is. In other words, the first network may be a genetic algorithm-based network that outputs task information for allocating tasks to a plurality of robots based on information representing the suitability of the behaviors of the plurality of robots in a predetermined situation, and the second network may be a convolutional neural network learned to output task information for allocating tasks to a plurality of robots based on information representing the suitability of the behaviors of the plurality of robots in a given situation.

According to an embodiment, the processor 110 may train the second network. According to an embodiment, the processor 110 checks the second information generated by processing the first information, obtains the checked second information as input information of the second network, and obtains the first network to which the first information is input. Obtain output information of the second network as target information for input information of the second network, and learn the second network based on the input information of the second network and the target information of the second network. According to another embodiment, the processor 110 checks second information generated by processing the first information, obtains the checked second information as input information of the second network, and performs tasks of experts corresponding to the second information. The information may be acquired as target information for the input information of the second network, and the second network may be learned based on the input information of the second network and the target information of the second network.

According to an embodiment, the plurality of robots may include individual robots, and the mission information may include mission values for allocating a mission to each robot. Specifically, the first task information identified based on the output of the first network includes a 1-1 task value for assigning a task to the first robot and a 1-2 task value for assigning a task to the second robot. , a 1-3 task value for assigning a task to a third robot, and a 1-4 task value for assigning a task to a fourth robot, and the second identified based on the output of the second network. The task information includes a 2-1 task value for assigning a task to the first robot, a 2-2 task value for assigning a task to the second robot, and a 2-3 task value for assigning a task to the third robot. value, and a 2nd-4th task value for allocating a task to the fourth robot.

5 shows a convolutional neural network as an example of the second network.

As shown in FIG. 5 , the convolutional neural network 500 may include convolutional layers, fully connected layers, and a softmax layer. According to an embodiment, the convolutional neural network 500 may include 5 convolutional layers, 2 fully connected layers, and a softmax layer. According to an embodiment, the convolutional neural network 500 outputs second information generated by processing from first information that is input information of the first network and output of the first network for the first information that is input information of the first network. It may be a neural network trained based on mission information for allocating missions to a plurality of robots. According to another embodiment, the convolutional neural network 500 is a neural network trained based on second information generated by processing first information, which is input information of the first network, and task information of an expert corresponding to the second information. can be Also, the convolutional neural network 500 may use a Flatten function, and the Flatten function may mean a function that changes the shape of data (tensor). For example, the Flatten function can convert 200x200x1 data into 40000x1 data.

According to an embodiment, the convolutional neural network 500 to which second information generated by processing the first information is input may output candidate task values through each of 40 neurons. The processor 110 sets the candidate task value having the highest output value of the softmax layer among 40 candidate task values as the 2-1 task value 510, the 2-2 task value 520, and the 2-3 task value ( 530), and the 2-4 task value 540. In other words, the processor 110 provides a 2-1 task value 510, a 2-2 task value 520, a 2-3 task value 530 as task information for allocating tasks to a plurality of robots, and the 2-4 task value 540 can be checked.

Also, the processor 110 may determine Cronbach's Alpha Coefficient of the output of the first network as reliability of the first network. Cronbach's alpha is a type of reliability index estimated based on the internal consistency of items. The Cronbach's alpha coefficient has a value of 0 to 1, and the closer to 1, the higher the reliability. Since the first network is a network based on a genetic algorithm, the processor 110 repeatedly performs a genetic algorithm on the first information, which is input information, collects output task values for each repetition, and cronizes the task values of all times. Baja alpha coefficient can be calculated. According to an embodiment, the processor 110 may determine whether the calculated Cronbach's alpha coefficient exists within a preset range, and if it exists within the preset range, the calculated Cronbach's alpha coefficient may be determined as reliability of the first network. . For example, when the preset range is 0.75 to 0.95, if the Cronbach's alpha coefficient is 0.8, the reliability of the first network is calculated as 0.8 because it is within the preset range, but if the Cronbach's alpha coefficient is 0.6, it is outside the preset range. The reliability of 1 network is not calculated as 0.6, and the processor 110 repeatedly performs the genetic algorithm until the Cronbach's alpha coefficient satisfies a preset range.

According to an embodiment, the processor 110 may calculate reliability of the second network based on outputs of the softmax layer of the convolutional neural network 500 . According to an embodiment, the processor 110 may calculate the reliability of the second network by calculating entropy of outputs of the softmax layer corresponding to the 40 neurons. In other words, the processor 110 may check the reliability of the 2-1 task value 510 as the reliability of the second network. According to another embodiment, the processor 110 may calculate the reliability of the second network by calculating entropy of outputs of the softmax layer corresponding to the other 40 neurons. In other words, the processor 110 may check the reliability of the 2-2nd task value 520, the 2-3rd task value 530, or the 2-4th task value 540 as the reliability of the second network. there is. According to another embodiment, the processor 110 provides reliability for the 2-1 task value 510, reliability for the 2-2 task value 520, and reliability for the 2-3 task value 530. , and the average value of the reliability for the 2-4 task value 540 is confirmed as the reliability of the second network, or the reliability for the 2-1 task value 510 and the 2-2 task value 520 The lowest value among the reliability of the second network, the reliability of the 2-3 task value 530, and the reliability of the 2-4 task value 540 may be identified as the reliability of the second network.

The processor 110 may calculate a first reliability for the first information from the first network to which the first information is input, and calculate a second reliability to the second information from the second network to which the second information is input. . Specifically, the processor 110 may calculate the first reliability based on the Cronbach's alpha coefficient of the output of the first network, and calculate the second reliability based on the outputs of the softmax layer of the second network.

According to an embodiment, the processor 110 may calculate the reliability of the second network according to Equations 1 and 2 below.

는 뉴럴 네트워크의 소프트맥스 레이어의 n개의 출력값들 중 i번째 출력값을 나타낸다. 따라서, 프로세서(110)는 수학식 1을 통해 엔트로피

를 계산하고, 수학식 2를 통해 엔트로피

의 역수로 신뢰도

를 계산할 수 있다. 여기서

는 0으로 나누는 것을 방지하기 위한 최소값으로 예를 들어 0.00001로 설정될 수 있다. 또한,

는 아래 수학식 3에 따라 계산될 수 있다. 수학식 3에서

는 소프트맥스 레이어에 입력되는 i번째 값을 나타낸다.In Equations 1 and 2,

represents an i-th output value among n output values of the softmax layer of the neural network. Therefore, the processor 110 calculates the entropy through Equation 1

Calculate , and entropy through Equation 2

reliability as the reciprocal of

can be calculated. here

may be set to, for example, 0.00001 as a minimum value for preventing division by zero. also,

Can be calculated according to Equation 3 below. in Equation 3

represents the i-th value input to the softmax layer.

를 계산할 수 있다.According to another embodiment, the processor 110 determines the reliability according to Equation 4 below.

can be calculated.

는 뉴럴 네트워크의 소프트맥스 레이어의 n개의 출력값들 중 가장 큰 값을 나타낸다. 수학식 4를 이용하여 신뢰도를 계산할 경우, 수학식 1 및 2를 이용하는 경우보다 연산 속도가 개선되는 효과를 가질 수 있다.In Equation 4,

represents the largest value among n output values of the softmax layer of the neural network. When the reliability is calculated using Equation 4, the calculation speed may be improved compared to the case using Equations 1 and 2.

, 제2-2 임무 값

, 제2-3 임무 값

, 및 제2-4 임무 값

을 확인할 수 있다(categorical mode).According to an embodiment, the processor 110 may assign a 2-1 task value for allocating tasks to a plurality of robots based on the output of the second network.

, the value of the 2-2 mission

, 2-3 mission value

, and the value of task 2-4

can be checked (categorical mode).

는 뉴럴 네트워크의 출력인 후보 제2-1 임무 값들 중에서 소프트맥스 레이어의 출력값인

가 가장 높은 i번째 후보 제2-1 임무 값을 나타내고, 제2-2 임무 값

는 뉴럴 네트워크의 출력인 후보 제2-2 임무 값들 중에서 소프트맥스 레이어의 출력값인

가 가장 높은 i번째 후보 제2-2 임무 값을 나타내고, 제2-3 임무 값

는 뉴럴 네트워크의 출력인 후보 제2-3 임무 값들 중에서 소프트맥스 레이어의 출력값인

가 가장 높은 i번째 후보 제2-3 임무 값을 나타내고, 제2-4 임무 값

는 뉴럴 네트워크의 출력인 후보 제2-4 임무 값들 중에서 소프트맥스 레이어의 출력값인

가 가장 높은 i번째 후보 제2-4 임무 값을 나타낸다.In Equation 5, the value of the 2-1 task

is the output value of the softmax layer among the candidate 2-1 task values that are outputs of the neural network.

represents the highest ith candidate 2-1 task value, and the 2-2 task value

is the output value of the softmax layer among the candidate 2-2 task values that are outputs of the neural network.

represents the highest ith candidate 2-2 task value, and the 2-3 task value

is the output value of the softmax layer among the candidate 2-3 task values that are outputs of the neural network.

represents the highest ith candidate 2-3 task value, and the 2-4 task value

is the output value of the softmax layer among the candidate task values 2-4, which are outputs of the neural network.

represents the highest i-th candidate task value 2-4.

6 shows an example of entropy for the outputs of the second network.

를 나타내고,

가 모두 균등하게 일정한 값들임을 나타낸다. 뉴럴 네트워크를 통해 출력된 확률 값들이 서로 비슷한 값을 가지면 엔트로피가 높게 계산되므로, 신뢰도는 반비례하여 낮아진다. 한편, 출력된 확률 값이 일부에 치우친 경우 엔트로피가 낮게 계산되므로, 신뢰도는 반비례하여 높아진다. 도 6을 참조하면, 좌측 그래프(610)의 경우 후보 임무 값들이 서로 비슷한 값을 가져 엔트로피는 2.9957로 비교적 큰 값으로 계산되고, 이는 제2 네트워크의 신뢰도가 낮음을 의미할 수 있다. 이와 같이, 인공신경망이 어려워하는 예시(예를 들어, 좌측 그래프(610))에 대한 출력은 일반적으로 확률 값이 한 곳으로 치우치지 않고, 여러 확률 값들이 서로 비슷한 값을 가지므로 신뢰도가 낮아진다.A graph 610 on the left represents output values of a softmax layer of a neural network according to an exemplary embodiment. Specifically, the graph 610 on the left is for the candidate 2-1 task values.

represents,

indicates that all are equally constant values. When probability values output through the neural network have similar values to each other, entropy is calculated to be high, and thus reliability decreases in inverse proportion. On the other hand, since entropy is calculated low when the output probability value is biased to a part, reliability increases in inverse proportion. Referring to FIG. 6 , in the case of the graph 610 on the left, the candidate task values have values similar to each other, so the entropy is calculated as a relatively large value of 2.9957, which may mean that the reliability of the second network is low. In this way, the probability value of the output for an example (eg, the graph 610 on the left) that is difficult for the artificial neural network is generally not biased, and since several probability values have values similar to each other, reliability is lowered.

를 나타내고, 특정 후보 제2-1 임무 값에 대한

가 높음을 나타낸다. 도 6을 참조하면, 우측 그래프(620)의 경우 후보 제2-1 임무 값이 일부에 치우쳐 엔트로피가 1.0457로 비교적 작은 값으로 계산되고, 이는 제2 네트워크의 신뢰도가 높음을 의미할 수 있다. 본원에서는 다채널 정보를 이용하여 각 뉴럴 네트워크로부터 계산되는 각 신뢰도의 융합을 통해 데이터의 양이 부족하면서 유사도가 적을 수 있는 전장 상황에서 다채널 정보의 상호 보완적 이용 방법을 개시한다.A graph 620 on the right represents output values of a softmax layer of a neural network according to another embodiment. Specifically, the graph 620 on the right is for the candidate 2-1 task values.

, and for a specific candidate 2-1 task value

indicates high. Referring to FIG. 6 , in the case of the graph 620 on the right, the value of the candidate 2-1 task is partially biased and the entropy is calculated as a relatively small value of 1.0457, which may mean that the reliability of the second network is high. In the present invention, a mutually complementary use method of multi-channel information is disclosed in a battlefield situation where the amount of data is insufficient and the degree of similarity may be low through convergence of each reliability calculated from each neural network using multi-channel information.

Referring back to FIG. 1 , the processor 110 may determine final task information for allocating tasks to a plurality of robots based on the first task information and the second task information. According to an embodiment, the processor 110 determines a first final task value from the 1-1 task value and the 2-1 task value, and determines a second final task value from the 1-2 task value and the 2-2 task value. The final task value is determined, the third final task value is determined from the first-third task value and the 2-3 task value, and the fourth final task value is determined from the first-4 task value and the 2-4 task value. can decide Subsequently, the processor 110 may allocate tasks to the plurality of robots based on the first final task value, the second final task value, the third final task value, and the fourth final task value. The processor 110 may transmit the determined final mission information to a plurality of robots through a communication device.

According to an embodiment, the processor 110 obtains the 1-1 task value and the 2-1 task value based on the first reliability and the second reliability determined based on the output of the first network and the output of the second network. A first final task value for assigning a task to the first robot may be determined, and the task is assigned to the second robot from the 1-2 task value and the 2-2 task value based on the first reliability and the second reliability. A third final task value for assigning a task to a third robot from the 1-3 task value and the 2-3 task value based on the first reliability and the second reliability , and a fourth final task value for assigning a task to the fourth robot may be determined from the 1-4 task values and the 2-4 task values based on the first reliability and the second reliability.

According to an embodiment, the first network expresses information representing the degree of fitness for the behavior of a plurality of robots in a 3-dimensional tensor in a predetermined situation, and assigns a task to a plurality of robots corresponding to the information representing the suitability. Mission information for outputting could be a network. The second network may be a network labeled with task information for allocating tasks to a plurality of robots corresponding to information representing the degree of fitness of a plurality of robots in a given situation in a matrix. Each network can receive input information and predict confirmation information (eg, task value). In a specific embodiment, the 1-1 task value, the 1-2 task value, the 1-3 task value, and the 1-4 task value from the first network and the 2-1 task value from the second network, A method of fusing the 2-2nd task value, the 2-3rd task value, and the 2-4th task value may be proposed. For example, it may be converged by averaging each task value obtained through the first network and the second network. According to an embodiment, when the respective reliability levels identified through each network are different, it may be more effective to fuse the output values of each network based on each reliability level rather than averaging. Hereinafter, as a specific embodiment, an example of weighting the output values (ie, task values) of each network based on the reliability of each network or determining the output value of a network with high reliability as the final output value will be described.

According to an embodiment, the processor 110 assigns a weight to each of the 1-1 task value and the 2-1 task value according to the ratio of the first reliability and the second reliability, and the weighted 1-1 The first final task value may be determined from the task value and the 2-1 task value. The processor 110 assigns a weight to each of the 1-2 task value and the 2-2 task value according to the ratio of the first reliability and the second reliability, and the weighted 1-2 task value and the 2- A second final task value may be determined from the 2 task values. The processor 110 assigns weights to the 1-3 task values and the 2-3 task values, respectively, according to the ratio of the first reliability and the second reliability, and the weighted 1-3 task values and the 2-3 task values are weighted. A third final task value may be determined from the three task values. The processor 110 assigns weights to the 1st-4th task values and the 2nd-4th task values, respectively, according to the ratio of the first reliability and the second reliability, and the weighted 1st-4th task values and the 2nd-4th task values are weighted. A fourth final task value may be determined from the 4 task values.

, 제2 최종 임무 값

, 제3 최종 임무 값

, 및 제4 최종 임무 값

을 결정할 수 있다.For example, the processor 110 calculates the first final task value according to Equation 6 below.

, the value of the second final task

, the value of the third final task

, and the value of the fourth final task

can determine

는 제1 신뢰도를 나타낼 수 있고,

는 제2 신뢰도를 나타낼 수 있다. 예를 들어, 제1 신뢰도는 제1 네트워크에 대한 신뢰도를 의미할 수 있고, 제2 신뢰도는 제2 네트워크에 대한 신뢰도를 의미할 수 있다. 또한, 수학식 6에서,

는 제1-1 임무 값을 나타낼 수 있고,

는 제2-1 임무 값을 나타낼 수 있고,

은 제1-2 임무 값을 나타낼 수 있고,

는 제2-2 임무 값을 나타낼 수 있고,

은 제1-3 임무 값을 나타낼 수 있고,

는 제2-3 임무 값을 나타낼 수 있고,

은 제1-4 임무 값을 나타낼 수 있고,

는 제2-4 임무 값을 나타낼 수 있다. 따라서, 프로세서(110)는 제1 신뢰도 및 제2 신뢰도의 비율에 따라 가중치가 부여된 제1-1 임무 값 및 제2-1 임무 값을 융합하여 제1 최종 임무 값을 결정할 수 있고, 제1 신뢰도 및 제2 신뢰도의 비율에 따라 가중치가 부여된 제1-2 임무 값 및 제2-2 임무 값을 융합하여 제2 최종 임무 값을 결정할 수 있고, 제1 신뢰도 및 제2 신뢰도의 비율에 따라 가중치가 부여된 제1-3 임무 값 및 제2-3 임무 값을 융합하여 제3 최종 임무 값을 결정할 수 있고, 제1 신뢰도 및 제2 신뢰도의 비율에 따라 가중치가 부여된 제1-4 임무 값 및 제2-4 임무 값을 융합하여 제4 최종 임무 값을 결정할 수 있다.In Equation 6,

May represent the first reliability,

may represent the second reliability. For example, the first reliability level may mean reliability of a first network, and the second reliability level may mean reliability of a second network. Also, in Equation 6,

May represent the 1-1 task value,

May represent the 2-1 task value,

May represent the 1st-2nd task value,

May represent the 2-2 task value,

May represent the 1-3 task values,

May represent the 2-3 task value,

May represent the 1-4 task values,

may represent the value of the 2-4 task. Accordingly, the processor 110 may determine the first final task value by fusing the 1-1 task value and the 2-1 task value weighted according to the ratio of the first reliability and the second reliability, and The second final task value may be determined by fusing the 1-2 task value and the 2-2 task value weighted according to the ratio of the reliability and the second reliability, and the second final task value may be determined according to the ratio of the first reliability and the second reliability. A third final task value may be determined by fusing the weighted task values 1-3 and task values 2-3, and tasks 1-4 weighted according to the ratio of the first reliability and the second reliability. The fourth final task value may be determined by fusing the value and the 2-4 task value.

According to another embodiment, the processor 110 may determine any one of the 1-1 task value and the 2-1 task value as the first final task value through a size comparison between the first reliability and the second reliability, Any one of the 1-2 task value and the 2-2 task value may be determined as the second final task value through the size comparison between the first reliability and the second reliability, and the size comparison between the first reliability and the second reliability may be determined. Through this, any one of the 1-3 task value and the 2-3 task value can be determined as the third final task value, and through the size comparison between the first reliability and the second reliability, the 1-4 task value and the 2- One of the four task values may be determined as the fourth final task value.

, 제2 최종 임무 값

, 제3 최종 임무 값

, 및 제4 최종 임무 값

을 결정할 수 있다.For example, the processor 110 calculates the first final task value according to Equation 7 below.

, the value of the second final task

, the value of the third final task

, and the value of the fourth final task

can determine

는 제1 신뢰도를 나타낼 수 있고,

는 제2 신뢰도를 나타낼 수 있다. 예를 들어, 제1 신뢰도는 제1 네트워크에 대한 신뢰도를 의미할 수 있고, 제2 신뢰도는 제2 네트워크에 대한 신뢰도를 의미할 수 있다. 또한 수학식 7에서

는 제1-1 임무 값을 나타낼 수 있고,

는 제2-1 임무 값을 나타낼 수 있고,

은 제1-2 임무 값을 나타낼 수 있고,

는 제2-2 임무 값을 나타낼 수 있고,

은 제1-3 임무 값을 나타낼 수 있고,

는 제2-3 임무 값을 나타낼 수 있고,

은 제1-4 임무 값을 나타낼 수 있고,

는 제2-4 임무 값을 나타낼 수 있다. 따라서, 프로세서(110)는 제1 신뢰도와 제2 신뢰도 간의 크기 비교에 따라, 신뢰도가 높은 네트워크로부터 획득된 제1-1 임무 값 또는 제2-1 임무 값을 제1 최종 임무 값으로 결정할 수 있고, 제1-2 임무 값 또는 제2-2 임무 값을 제2 최종 임무 값으로 결정할 수 있고, 제1-3 임무 값 또는 제2-3 임무 값을 제3 최종 임무 값으로 결정할 수 있고, 제1-4 임무 값 또는 제2-4 임무 값을 제4 최종 임무 값으로 결정할 수 있다.In Equation 7,

May represent the first reliability,

may represent the second reliability. For example, the first reliability level may mean reliability of a first network, and the second reliability level may mean reliability of a second network. Also in Equation 7

May represent the 1-1 task value,

May represent the 2-1 task value,

May represent the 1st-2nd task value,

May represent the 2-2 task value,

May represent the 1-3 task values,

May represent the 2-3 task value,

May represent the 1-4 task values,

may represent the value of the 2-4 task. Accordingly, the processor 110 may determine the 1-1 task value or the 2-1 task value obtained from the high-reliability network as the first final task value according to the size comparison between the first reliability and the second reliability, , the 1-2 task value or the 2-2 task value may be determined as the second final task value, the 1-3 task value or the 2-3 task value may be determined as the third final task value, The 1-4 task value or the 2-4 task value may be determined as the fourth final task value.

According to an embodiment, the plurality of robots include a first robot, a second robot, a third robot, and a fourth robot, and the second reliability is a 2-1 reliability for a 2-1 task value, a second- 2-2 reliability for the value of task 2, 2-3 reliability for the value of task 2-3, and 2-4 reliability for the value of task 2-4 may be included. The processor 110 determines the first final task value from the 1-1 task value and the 2-1 task value based on the first reliability and the 2-1 task value, and determines the first and 2-2 task values. Based on the 1-2 task value and the 2-2 task value, the second final task value is determined, and the 1-3 task value and the 2-3 task value are based on the first reliability and the 2-3 task value. A third final task value may be determined from , and a fourth final task value may be determined from the first to fourth task values and the second to fourth task values based on the first reliability and the 2 to 4th reliability. According to an embodiment, the processor 110 assigns a weight to each of the 1-1 task value and the 2-1 task value according to the ratio of the 1st reliability level and the 2-1st reliability level, and the weighted first The first final task value may be determined from the -1 task value and the 2-1 task value. In addition, the processor 110 assigns weights to the 1-2 task values and the 2-2 task values, respectively, according to the ratio of the 1st reliability and the 2-2 task values, and the weighted 1-2 task values. and the second final task value may be determined from the value of the 2-2 task. Similarly, the processor 110 assigns weights to the 1-3 task values and the 2-3 task values, respectively, according to the ratio of the first reliability and the 2-3 task values, and the weighted 1-3 task values and A third final task value may be determined from the 2-3 task values, weights are assigned to each of the 1-4 task values and the 2-4 task values according to the ratio of the first reliability and the 2-4 task values, A fourth final task value may be determined from the weighted first-four task values and second-four task values. According to another embodiment, the processor 110 may determine any one of the 1-1 task value and the 2-1 task value as the first final task value through a size comparison between the 1st reliability level and the 2-1 level reliability level. and, through a comparison of magnitudes between the first reliability and the 2-2 reliability, any one of the 1-2 task value and the 2-2 task value may be determined as the second final task value, and the first reliability and the second-2 task value may be determined. Any one of the 1-3 task values and the 2-3 task values may be determined as the third final task value through the size comparison between the 3 reliability values, and the first and 2-4 task value comparisons may determine the first task value. Any one of the -4 task value and the 2-4 task value may be determined as the fourth final task value.

According to an embodiment, the electronic device 100 may obtain a first final task value, a second final task value, a third final task value, and a fourth final task value through a first network based on a genetic algorithm and a second network based on a neural network. A final task value may be determined, and the first robot, the second robot, the third robot, and the second final task value may be determined based on the first final task value, the second final task value, the third final task value, and the fourth final task value. 4 You can assign tasks to robots. The electronic device 100 allocates the first robot, the second robot, the third robot, and the fourth robot through the first final task value, the second final task value, the third final task value, and the fourth final task value. Information about one task can be transmitted to each robot. The electronic device 100 complementarily uses a network based on a neural network as well as a network based on a genetic algorithm, thereby improving the disadvantages of a genetic algorithm that requires a large amount of data and unlearned data having a relatively low degree of similarity. , it is possible to improve the task assignment performance by improving the disadvantages of the neural network, which has poor output quality. In addition, by using a neural network, the electronic device 100 can infer a task value faster than when only a genetic algorithm-based network is used, and assign tasks to multiple robots.

7 illustrates an embodiment in which a processor determines a first final task value, a second final task value, a third final task value, and a fourth final task value.

Referring to FIG. 7 , the processor 110 performs the first-1st operation based on the output of the first network 730 to which the first information 710 representing the degree of fitness for the behavior of a plurality of robots is input in a predetermined situation. It is possible to check the task value, the 1-2 task value, the 1-3 task value, and the 1-4 task value, the 1-1 task value, the 1-2 task value, the 1-3 task value, And it is possible to check the first reliability for the 1-4 task values. In addition, the processor 110 performs the 2-1 task value, the second information 720 indicating the degree of suitability for the behavior of a plurality of robots in a predetermined situation, respectively, based on the output of the second network 740 input thereto. It is possible to check the 2-2 task value, the 2-3 task value, and the 2-4 task value, the 2-1 reliability for the 2-1 task value, and the 2-1 reliability for the 2-2 task value. 2 reliability, 2-3 reliability for the 2-3 task value, and 2-4 reliability for the 2-4 task value can be confirmed.

According to an embodiment, the processor 110 determines a first final task value from the 1-1 task value and the 2-1 task value based on the first reliability and the 2-1 task value, and determines the first final task value and Based on the 2-2 reliability, the second final task value is determined from the 1-2 task value and the 2-2 task value, and the 1-3 task value and A third final task value may be determined from the 2-3 task value, and a fourth final task value may be determined from the 1-4 task value and the 2-4 task value based on the first reliability and the 2-4 task value. there is. For example, the processor 110 may determine the 1-1st task value and the 2-1st task value according to the ratio of the 1st reliability and the 2-1st reliability or the size comparison between the 1st reliability and the 2-1st reliability. Determines a first final task value for assigning a task to the first robot, and according to a ratio of the first reliability and the 2-2 reliability or a size comparison between the first reliability and the 2-2 reliability, the 1-2 task A second final task value for assigning a task to the second robot is determined from the value and the 2-2 task value, and the ratio between the first reliability and the 2-3 reliability or the size between the first reliability and the 2-3 reliability According to the comparison, a third final task value for assigning a task to a third robot is determined from the 1-3 task values and the 2-3 task values, and the ratio of the first reliability and the 2-4 reliability or the first According to the size comparison between the reliability and the 2-4th reliability, a fourth final task value for assigning a task to the fourth robot may be determined from the 1-4th task value and the 2-4th task value. According to an embodiment, the final task value may be a value corresponding to a predefined task. The processor 110 assigns each of the tasks corresponding to the first final task value, the second final task value, the third final task value, and the fourth final task value to the first robot, the second robot, the third robot, and the second final task value. 4 can be assigned to robots.

8 shows an embodiment in which an electronic device operates.

In step S810 for data acquisition (Data Acquisition), the electronic device 100 may acquire information representing the degree of suitability for the behavior of the robot in a predetermined situation. For example, the electronic device 100 may obtain an information tensor as first information and obtain 2D image information obtained by forming an information matrix obtained by processing the first information as second information. According to an embodiment, the electronic device 100 may obtain first information and second information each representing a degree of suitability for a robot's behavior in an battlefield situation stored in the memory 120 .

In the step for data editing (S820), the electronic device 100 edits the data obtained in S810 so that a portion not required for repetition of the first network or a portion not required for learning of the second network is edited. can be removed. For example, when the second information representing the suitability of a robot's behavior in a battlefield situation previously acquired to learn or train a second network is duplicated due to a simple human error, the electronic device 100 provides redundant data. can be removed.

In step S830 for training, the electronic device 100 may train the second network based on the data edited in S820. Specifically, the electronic device 100 may train or learn the second network through second information, which is input information, and first task information, which is an output of the first network, which is a label, or task information of an expert.

In the step for testing (S840), the electronic device 100 may check second task information for allocating tasks to a plurality of robots using the second network learned or trained in S830. Depending on the mission information, final mission information can be determined and provided. Specifically, the electronic device 100 determines a first final task value, a second final task value, a third final task value, and a fourth final task value based on the first information and the second information, and determines the first final task value. The final task value, the second final task value, the third final task value, and the fourth final task value may be provided to each robot.

According to an embodiment, the electronic device 100 may transmit final mission information to a plurality of robots. If the outputted task assignment information is extremely different from the task information of the actual skilled worker or the task information output from the first network, it may be determined that the electronic device 100 has not accurately provided the tasks to the plurality of robots. If the tasks are not accurately provided to the plurality of robots, the electronic device 100 may relearn the second network.

9 illustrates a method of operating an electronic device according to an exemplary embodiment.

Since each step of the operating method of FIG. 9 can be performed by the electronic device 100 of FIG. 1 , descriptions of overlapping contents with those of FIG. 9 will be omitted.

In step S910, the electronic device 100 may obtain first information and second information each representing a degree of suitability for a robot's behavior in a predetermined situation. The first information may include information in the form of a 3D tensor, and the second information may include 2D image information. The first information and the second information may be generated based on resource information of each of a plurality of robots, battlefield situation information, information about a fitness function for each battlefield situation, and resource-robot behavior matching condition information.

In step S920, the electronic device 100 checks first task information for allocating tasks to a plurality of robots based on the output of the first network to which the first information is input, and the second network to which the second information is input. Second task information for allocating tasks to a plurality of robots may be checked based on the output of . The first network is a genetic algorithm-based network that outputs mission information for allocating missions to a plurality of robots based on information representing the suitability of robot behavior in a battlefield situation, and the second network is a network based on robot behavior in a battlefield situation. It may be a convolutional neural network learned to output task information for allocating tasks to a plurality of robots based on information representing the degree of suitability for .

In step S930, the electronic device 100 may determine final task information for allocating tasks to a plurality of robots based on the first task information and the second task information. The electronic device 100 determines a first final task value from the 1-1 task value and the 2-1 task value, and determines a second final task value from the 1-2 task value and the 2-2 task value. and determine the third final task value from the 1-3 task values and the 2-3 task values, and determine the 4th final task value from the 1-4 task values and the 2-4 task values. Also, the electronic device 100 determines the first final task value from the 1-1 task value and the 2-1 task value based on the first reliability and the second reliability, and determines the 1-2 task value and the second task value. Determine the second final task value from the -2 task value, determine the third final task value from the 1-3 task value and the 2-3 task value, and determine the 1-4 task value and the 2-4 task value It is possible to determine the fourth final task value from

In step S940, the electronic device 100 may provide final mission information. The electronic device 100 may provide the determined first final task value, second final task value, third final task value, and fourth final task value. Also, the electronic device 100 may transmit final mission information to a plurality of robots through a communication device.

An electronic device according to the above-described embodiments includes a processor, a memory for storing and executing program data, a permanent storage unit such as a disk drive, a communication port for communicating with an external device, a touch panel, a key, User interface devices such as buttons and the like may be included. Methods implemented as software modules or algorithms may be stored on a computer-readable recording medium as computer-readable codes or program instructions executable on the processor. Here, the computer-readable recording medium includes magnetic storage media (e.g., read-only memory (ROM), random-access memory (RAM), floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM) ), and DVD (Digital Versatile Disc). A computer-readable recording medium may be distributed among computer systems connected through a network, and computer-readable codes may be stored and executed in a distributed manner. The medium may be readable by a computer, stored in a memory, and executed by a processor.

This embodiment can be presented as functional block structures and various processing steps. These functional blocks may be implemented with any number of hardware or/and software components that perform specific functions. For example, embodiments may include integrated circuit configurations such as memory, processing, logic, look-up tables, etc., that may execute various functions by means of the control of one or more microprocessors or other control devices. can employ them. Similar to components that can be implemented as software programming or software elements, the present embodiments include data structures, processes, routines, or various algorithms implemented as combinations of other programming constructs, such as C, C++, Java ( It can be implemented in a programming or scripting language such as Java), assembler, or the like. Functional aspects may be implemented in an algorithm running on one or more processors. In addition, this embodiment may employ conventional techniques for electronic environment setting, signal processing, and/or data processing. Terms such as “mechanism”, “element”, “means” and “composition” may be used broadly and are not limited to mechanical and physical components. The term may include a meaning of a series of software routines in association with a processor or the like.

The foregoing embodiments are merely examples and other embodiments may be implemented within the scope of the claims described below.

Claims (12)

A method in which an electronic device assigns a task to a plurality of robots,
First information including information representing the degree of fitness for the behavior of the plurality of robots in the form of a 3-dimensional tensor in a predetermined situation, and the degree of fitness for the behavior of the plurality of robots in the predetermined situation as a two-dimensional image obtaining second information including the information indicated by ;
Check first task information for allocating tasks to the plurality of robots, which is an output of the first network to which the first information is input, and tasks to the plurality of robots, which are output of a second network to which the second information is input checking second task information for allocating;
determining final task information for allocating tasks to the plurality of robots based on the first task information and the second task information; and
Including providing the final mission information,
The first network,
A network based on a genetic algorithm that outputs mission information for allocating missions to a plurality of robots based on information representing the degree of fitness for robot behavior in a given situation,
The second network,
A method, which is a convolutional neural network learned to output task information for allocating tasks to a plurality of robots based on information representing the degree of suitability of the robot's behavior in a given situation. delete According to claim 1,
The obtaining step is
The first information and the second information are obtained based on the resource information of each of the plurality of robots, the information about the predetermined situation, the information about the fitness function for each predetermined situation, and the information about the resource-robot behavior matching condition. A method comprising generating. delete According to claim 1,
The second information is information generated by processing from the first information,
acquiring the second information as input information of the second network and acquiring output information of the first network into which the first information is input as target information for the input information of the second network; and
and learning the second network based on the input information and the target information. According to claim 1,
acquiring the second information as input information of the second network and acquiring task information of an expert corresponding to the second information as target information for the input information of the second network; and
and learning the second network based on the input information and the target information. According to claim 1,
The plurality of robots include a first robot, a second robot, a third robot, and a fourth robot,
The first mission information,
A 1-1 task value for allocating a task to the first robot, a 1-2 task value for allocating a task to the second robot, and a 1-3 task value for allocating a task to the third robot , and 1-4 task values for allocating tasks to the fourth robot,
The second task information,
A 2-1 task value for allocating a task to the first robot, a 2-2 task value for allocating a task to the second robot, and a 2-3 task value for allocating a task to the third robot , and a 2-4 task value for assigning a task to the fourth robot,
The final mission information,
A first final task value for assigning a task to the first robot, a second final task value for assigning a task to the second robot, a third final task value for assigning a task to the third robot, and the A fourth final task value for assigning a task to a fourth robot;
The determining step is
The first final task value is determined from the 1-1 task value and the 2-1 task value, and the second final task value is determined from the 1-2 task value and the 2-2 task value and determining the third final task value from the 1-3 task value and the 2-3 task value, and the 4th final task value from the 1-4 task value and the 2-4 task value. A method comprising the step of determining According to claim 1,
The checking step is
The first reliability, the 1-1 task value, the 1-2 task value, the 1-3 task value, and the 1-4 task value are checked based on the output of the first network to which the first information is input. And based on the output of the second network to which the second information is input, the second reliability, the 2-1st task value, the 2-2nd task value, the 2-3rd task value, and the 2-4th task value Including the step of checking,
The determining step is
Based on the first reliability and the second reliability, a first final task value is determined from the 1-1 task value and the 2-1 task value, and the 1-2 task value and the 2-1 task value are determined. A second final task value is determined from the 2 task value, a third final task value is determined from the 1-3 task value and the 2-3 task value, and the 1-4 task value and the 2-3 task value are determined. determining a fourth final task value from the four task values. According to claim 8,
The first reliability is,
Calculated based on Cronbach's Alpha Coefficient of the output of the first network,
The second reliability is,
calculated based on outputs of the softmax layer of the second network. According to claim 1,
and transmitting the final task information to the plurality of robots via a communication device. As an electronic device that assigns tasks to a plurality of robots,
a memory in which at least one program is stored; and
By executing the at least one program, first information including information representing the degree of fitness for the behavior of the plurality of robots in the form of a three-dimensional tensor in a predetermined situation and the plurality of robots in the predetermined situation Obtaining second information including information representing the degree of fitness for the behavior of in a two-dimensional image;
Check first task information for allocating tasks to the plurality of robots, which is an output of the first network to which the first information is input, and tasks to the plurality of robots, which are output of a second network to which the second information is input Check the second mission information for allocating,
Determine final task information for allocating tasks to the plurality of robots based on the first task information and the second task information, and
Including a processor providing the final task information,
The first network,
A network based on a genetic algorithm that outputs mission information for allocating missions to a plurality of robots based on information representing the degree of fitness for robot behavior in a given situation,
The second network,
An electronic device that is a Convolutional Neural Network that has been trained to output task information for allocating tasks to a plurality of robots based on information representing a degree of suitability for a robot's behavior in a given situation. A computer-readable non-transitory recording medium recording a program for executing a method of assigning a task to a plurality of robots on a computer,
The method,
First information including information representing the degree of fitness for the behavior of the plurality of robots in the form of a 3-dimensional tensor in a predetermined situation, and the degree of fitness for the behavior of the plurality of robots in the predetermined situation as a two-dimensional image obtaining second information including the information indicated by ;
Check first task information for allocating tasks to the plurality of robots, which is an output of the first network to which the first information is input, and tasks to the plurality of robots, which are output of a second network to which the second information is input checking second task information for allocating;
determining final task information for allocating tasks to the plurality of robots based on the first task information and the second task information; and
Including providing the final mission information,
The first network,
A network based on a genetic algorithm that outputs mission information for allocating missions to a plurality of robots based on information representing the degree of fitness for robot behavior in a given situation,
The second network,
A non-transitory recording medium, which is a convolutional neural network trained to output task information for allocating tasks to a plurality of robots based on information representing the degree of suitability of the robot's behavior in a given situation.

Images