JP6739972B2 - Multi-agent system, task allocation device, task allocation method and program - Google Patents

Description

Embodiments of the present invention relate to a multi-agent system, a task allocation device, a task allocation method and a program applied to the system.

In recent years, attention has been paid to the use of a multi-agent system in which a plurality of agents communicate with each other via a network and cooperate to execute various tasks (processes). For example, in a control system for controlling a plurality of drones (drone: unmanned aerial vehicle), a multi-agent system can be applied to realize a cooperation function for controlling each drone in cooperation.

Generally, an agent senses a situation (environment), recognizes a task to be executed based on the sensing information, and autonomously uses a resource (computer resource) to execute the task.

Here, the multi-agent system includes a task allocation (distribution) function that allocates (distributes) a task for controlling each drone to each agent through communication between the agents. As a method of implementing the task assignment function, there is a method (contract net protocol) in which each agent performs task exchange by executing information exchange using a communication function via a network.

JP, 2007-52683, A JP 2002-228396A International publication WO2006/118193 JP, 2009-55401, A

In the multi-agent system, the task assignment function is realized by the communication between the agents. When the communication between the agents becomes unstable, there may occur a state in which communication is temporarily unavailable or a state in which communication time increases and communication efficiency decreases.

In such a case, task allocation cannot be performed, such as a situation in which multiple agents are in charge of a single task, or the agent in charge of the task cannot be specified, or a fixed time (effective time) It may not be possible to execute within.

Therefore, there is a problem of realizing a multi-agent system that can appropriately execute task assignment even when communication between agents becomes unstable.

In the present embodiment, in a multi-agent system in which a plurality of agents are connected to a network and communicate with each other, each of the agents includes a task recognition means, a task allocation means, and a prediction means. The task recognition means recognizes a task to be executed. The task assigning unit selects a first agent capable of assigning the task from other agents based on the recognition result of the task recognizing unit, and when communication with the first agent is possible. , Assigning the task to the first agent. When the communication with the first agent is impossible, the predicting unit predicts a second agent that is highly likely to allocate the task to the first agent from another agent. ..

FIG. 3 is a block diagram for explaining the configuration of a multi-agent system according to the embodiment. 3 is a flowchart for explaining the operation of the multi-agent system according to the embodiment. The figure which shows an example of the network map information regarding embodiment. The figure which shows an example of the other agent task recognition performance information regarding embodiment. The figure which shows an example of the other agent task allocation rule regarding embodiment. FIG. 6 is a schematic diagram for explaining an example of task assignment processing according to the embodiment. FIG. 6 is a schematic diagram for explaining an example of task assignment processing according to the embodiment.

Embodiments will be described below with reference to the drawings.
[System configuration]
FIG. 1 is a diagram showing the configuration of a multi-agent system according to this embodiment. In this system, each of the agents 10-1 to 10-N communicates with each other in a plurality of computers 1-1 to 1-N connected to the network 2 and includes a task allocation (task distribution) function described later. is there.

The multi-agent system according to this embodiment is applied to, for example, a control system that controls a plurality of drones, and each agent 10-1 to 10-N controls each drone in cooperation with each other. The agents 10-1 to 10-N may be referred to as agents A, B, C, D, etc. for convenience. That is, when the agent 10-1 corresponds to the agent A (parent agent), the other plurality of agents (other agents) 10-N correspond to the agents B, C, and D.

In the present embodiment, as shown in FIG. 1, the agents 10-1 to 10-N have basically the same internal configuration. Hereinafter, the internal configuration will be described with the agent 10-1 as a representative, and the description of the internal configuration of other agents will be omitted.

The agent 10-1 has a configuration including software modules and databases (information) 15, 16, 18 to 20 for realizing the respective functions 11 to 14 and 17. The agent 10-1 includes a task allocation unit 11, a task recognition unit 12, a sensing unit 13, a task execution unit 14, and another agent task allocation prediction unit (hereinafter, referred to as a task allocation prediction unit) 17.

The task allocation unit 11 refers to the task allocation rule database (hereinafter, the database will be referred to as DB) 15 and the other agent task allocation DB 16 for the task to be executed (for convenience, the task X), and the other agent 10- The task allocation process of allocating (distributing) the task X to N is executed.

The task allocation rule DB 15 is information indicating a task allocation rule (one of the task allocation rules described in FIG. 5) when the agent 10-1 (agent A) executes the task allocation process. The other agent task assignment DB 16 is history information when performing task assignment processing based on the so-called contract net protocol. This history information is information on task allocation processing when information is exchanged with another agent by communication.

The task recognition unit 12 recognizes the task X to be executed based on the sensing information of the sensing unit 13, and notifies the task allocation unit 11 of it. The sensing unit 13 senses (detects) a situation (environment) when a multi-agent system is applied, for example, when controlling a plurality of drones in cooperation with each other. As a specific example, the sensing unit 13 communicates with, for example, the radar 30-1, and executes a predetermined detection process (for example, flight conditions such as positions of a plurality of drones) according to a detection signal from the radar 30-1. To do. The task recognition unit 12 recognizes, for example, a task X for controlling a designated drone based on the sensing information.

The task execution unit 14 executes the task assigned by the task assignment unit 11. Here, the agent 10-1 (agent A) executes the task assigned to the agent A or the task assigned from the other agent 10-N by the task assigning unit 11. As a specific example, the task execution unit 14 communicates with the controller 31-1 that controls each drone in the drone control system, and causes the controller 31-1 to execute the assigned task (for example, control process). The controller 31-1 may be the actuator of the device to be controlled.

The task allocation prediction unit 17 cooperates with the task allocation unit 11 to predict from another agent 10-N an agent that may allocate the task X to the corresponding agent (here, B). The agent B is an agent to which the task X is assigned by the normal task assigning process of the task assigning unit 11, and is in a state where communication with the agent A is not possible thereafter.

As will be described later, the task allocation prediction unit 17 refers to the network map DB 18, other agent task recognition performance DB 19, and other agent task allocation rule DB 20, and executes the prediction process. The agents 10-1 to 10-N need not share the other agent task allocation rule DB 20 as long as the task allocation rules of the task allocation rule DB 15 have the same contents.
[Task assignment process]
FIG. 2 is a flowchart for explaining the task allocation processing of this embodiment. The task assignment process of this embodiment will be described with reference to the flowchart of FIG.

In the present embodiment, for convenience, the agent 10-1 is the agent A and the other agents 10-N are the agents B, C, D and the like. Agent A is a parent agent that assigns tasks to other agents B, C, D, and so on.

As shown in FIG. 2, in the agent A, the task recognition unit 12 recognizes the necessity of executing the task X (in other words, the task X to be executed) based on the sensing information of the sensing unit 13 (step S1). .. Specifically, the sensing information is, for example, information indicating flight conditions such as positions of a plurality of drones detected according to the detection signal from the radar 30-1.

When the task assignment unit 11 refers to the other agent task assignment DB 16 and recognizes that the assignment process of the task X has already been started among the other agents B, C, D, etc., the task assignment process ends. (YES in step S2). Here, the agent A holds the history information when the other agents B, C, D, etc. execute the task allocation processing by exchanging information in the communication between the agents, as the other agent task allocation DB 16. That is, the history information is information when performing the task allocation process based on the so-called contract net protocol.

When the task X allocation process is not started by another agent (NO in step S2), the task allocation unit 11 may be in charge of the task X to the other agents B, C, D, etc. by communication between the agents. (In other words, the possibility of executing task X) is inquired (step S3). When no response is received from any of the agents in response to this inquiry (inquiry) (NO in step S4), the task allocation unit 11 terminates the task allocation processing for the other agents B, C, D, and The task X is executed by its own task execution unit 14. Note that the task allocation unit 11 may execute processing other than that executed by its own task execution unit 14.

On the other hand, when there is another agent that responds to the inquiry within a fixed time (valid time) (YES in step S4), the task allocation unit 11 refers to the task allocation rule DB 15 and sets the task allocation rule set in advance. Based on this, the agent B determined to be the most suitable for allocation is selected (step S5). A specific example is a task allocation rule as shown in FIG.

When the task allocator 11 can communicate with the selected agent B within a fixed time, that is, if the communication is maintained (YES in step S6), the task B allocation notification is sent to the agent B. Is performed (step S7). As a result, the agent B executes the task X assigned by the agent A by the task execution unit 14. As a specific example, the agent B communicates with the controller 31-N that controls the drone, for example, and causes the assigned task (for example, control process) X to be executed.

Here, the task allocation unit 11 cooperates with the task allocation prediction unit 17 when communication with the selected agent B is impossible within a certain time, that is, communication is unstable (NO in step S6). The task allocation prediction unit 17 refers to the network map DB 18, the other agent task recognition performance DB 19, and the other agent task allocation rule DB 20 to perform a prediction process of predicting whether or not another agent allocates the task X to the agent B. Execute (step S8).

Specifically, the task allocation prediction unit 17 compares the multiplication result of the probability of each condition with the reference value based on the network map information, the other agent task recognition performance information, and the other agent task allocation rule, and determines the reference value. It is predicted that the exceeding agent C is likely to assign the task X to the agent B. In short, the task allocation prediction unit 17 predicts that the agent C has a high possibility of allocating the task X to the agent B based on the probability information indicating the possibility of allocating the task.

Here, the network map DB 18 includes the network map information as shown in FIG. 3, and is specifically information on the communication state of each arc (communication connection between agents). In particular, the connection probability varies greatly when the communication between agents is wireless communication. This network map information is periodically acquired by exchanging information between agents. The other agent task recognition performance DB 19 includes other agent task recognition performance information as shown in FIG. This information is information that depends on the sensing performance, and is based on, for example, the installation position of the radar 30-N and the detection distance. The other agent task assignment rule DB 20 includes task assignment rules as shown in FIG. If the task allocation rules of the task allocation rule DB 15 of each agent have the same content, it is not necessary to share the other agent task allocation rule DB 20.

Returning to FIG. 2, when the task allocation predicting unit 17 predicts that the agent C communicates with the agent B within a fixed time to allocate the task X to the agent B (YES in step S9). , Task allocation unit 11 is notified. As a result, the task allocation unit 11 ends the task allocation process.

When the task allocation predicting unit 17 cannot predict an agent with a high possibility of allocating the task X to the agent B (NO in step S9), the task allocation unit 11 can again communicate with the agents other than the agent B within a fixed time. Select another agent. That is, the task allocation unit 11 refers to the task allocation rule DB 15 and selects the agent D that is determined to be the optimum allocation target based on the preset task allocation rule (step S10).

The task allocation unit 11 notifies the selected agent D of the allocation of the task X to the agent D when the communication is possible within a certain time, that is, when the communication is maintained (step S4). S11). As a result, the agent D executes the task X assigned by the agent A by the task execution unit 14.

With regard to the task assignment processing of the present embodiment as described above, the operation and effect will be described with reference to the schematic diagrams of FIGS. 6 and 7.

First, the agent A selects the agent B as the assignment target of the task X, and if the communication between the agents A and B is possible, the agent A executes the assignment process of the task X to the agent B and ends the process. Here, as shown in FIG. 6, when a state (60) in which communication between the agents A and B is not possible occurs, the task allocation prediction unit 17 causes the agent C to communicate with the agent B within a fixed time (effective time). (61) If it is predicted that the task X is likely to be allocated, the task allocation unit 11 ends without executing anything. Here, the agent A is in a state where it cannot communicate with the predicted agent C.

That is, when the communication between the agents A and B is unstable, and the agent A predicts the agent C having a high possibility of allocating the task X to the agent B, the task allocation unit 11 terminates without executing anything. Becomes

Therefore, it is possible to avoid a situation in which a plurality of agents are in charge of the same task X and the computer resources are wasted. Further, when a plurality of agents are in charge of the same task X, it becomes impossible for each agent to allocate resources to the execution of other tasks, but such a situation can be avoided. As a result, task allocation (task distribution) can be appropriately executed even if the communication between the agents becomes unstable.

Further, as shown in FIG. 7, in a state (62) in which the agents B and C cannot communicate with each other, if the agent C having a high possibility of assigning the task X to the agent B within a certain time cannot be predicted, the task assigning unit 11 Executes the process of selecting the agent D as the assignment target of the task X again. If the agent A can communicate with the agents A and D within a certain time (effective time), the agent A executes the assignment process of the task X to the agent D and ends the process. Therefore, even when the communication between the agents has become unstable, it is possible to avoid the situation where the agent who can handle the task X cannot be specified.

In the present embodiment, the task allocation prediction unit 17 and the DBs 18-20 have been described as being provided in each agent, but may be provided only in the agent A.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the scope of equivalents thereof.

1-1 to 1-N... Computer, 2... Network, 10-1 to 10-N... Agent,
11... Task allocation unit, 12... Task recognition unit, 13... Sensing unit,
14... Task execution unit, 15... Task assignment rule database (DB),
16... Other agent task allocation DB, 17... Other agent task allocation prediction unit,
18... network map DB, 19... other agent task recognition performance DB,
20... Other agent task assignment rule DB, 30-1 to 30-N... Radar,
31-1 to 31-N... Controller.

Claims (14)

In a multi-agent system where multiple agents connect to the network and communicate,
Each of the agents is
Task recognition means for recognizing tasks to be executed,
Based on the recognition result of said task recognition means, among the plurality of agents, the agents of oneself and others, selects the first agent assignments are possible the tasks, which can communicate with the first agent In this case, a task allocation unit that executes the task allocation process for the first agent,
If not possible to communicate with the first agent, from said host other agents, and predicting means for possibility of performing the assignment of the task to the first agent to predict high second agent A multi-agent system that includes: The task allocation means,
The multi-agent according to claim 1, wherein, when the second agent cannot be predicted by the predicting means, a third agent which can communicate with the other agent and which can allocate the task is selected from the other agents. system. The task allocation means,
The multi-agent system according to claim 1, wherein when the second agent is predicted by the prediction means, the task allocation process is terminated. The task allocation means,
By communication with a plurality of agent of the own and other, selects the first agent from the agent that responds to queries whether it is possible to allocation of the task,
The multi-agent system according to claim 1, wherein when there is no response to the inquiry, the task allocation process is terminated. 5. The predicting means predicts a second agent that is highly likely to assign the task to the first agent, based on probability information including a communication connection probability between agents. The multi-agent system according to any one of items. 5. The predicting means predicts a second agent, which is highly likely to allocate the task to the first agent within a fixed time, based on the communication speed between the agents. The multi-agent system according to item 1. 5. The predicting means predicts a second agent that is highly likely to allocate the task to the first agent, based on probability information including a task recognition probability of another agent. The multi-agent system according to any one of items. 5. The predicting means predicts a second agent that is highly likely to allocate the task to the first agent based on a task allocation rule of another agent. The multi-agent system described in. 5. The predicting means predicts a second agent that is highly likely to allocate the task to the first agent based on a task allocation rule common to all agents. The multi-agent system described in Section. A task assignment device applied to a multi-agent system in which a plurality of agents connect to and communicate with a network,
The task assignment device is provided in one agent or all of a plurality of agents,
Task recognition means for recognizing tasks to be executed,
Based on the recognition result of said task recognition means, among the plurality of agents, and selects the first agent assignments are possible the task from the agent of oneself and others, if possible to communicate with the first agent A task assigning means for executing the task assigning process for the first agent;
If not possible to communicate with the first agent, from said host other agents, and predicting means for possibility of performing the assignment of the task to the first agent to predict high second agent A task allocation device comprising: The task allocation device according to claim 10, wherein the prediction unit predicts a second agent that is highly likely to allocate the task to the first agent based on a task allocation rule common to all agents. .. It said predicting means, based on the probability information including a communication connection probabilities between agents, according to claim 10, possibility of performing assignment of the task to the first agent to predict high second agent Task assignment device. A task allocation method applied to a multi-agent system in which multiple agents connect to and communicate with a network,
The process of recognizing the tasks to be performed,
A process of selecting a first agent capable of assigning the task from other agents of the plurality of agents based on the recognition result of the task;
A process of executing the task assignment process for the first agent when communication with the first agent is possible;
If not possible to communicate with the first agent, the self from another agent, and a process of predicting the second agent is likely to perform assignment of the task for the first agent A task allocation method provided. Applied to a multi-agent system in which a plurality of agents connect to and communicate with a network, in a computer incorporating the plurality of agents,
The process of recognizing the tasks to be performed,
A process of selecting a first agent capable of assigning the task from other agents of the plurality of agents based on the recognition result of the task;
A process of executing the task assignment process for the first agent when communication with the first agent is possible;
If not possible to communicate with the first agent, the self from another agent, and a process of predicting the second agent is likely to perform assignment of the task for the first agent A program that causes the computer to execute a procedure that the computer has.