Control apparatus, control system, control method, and program

A control system configured to control a plurality of agents in a distributed manner, each of the plurality of agents being configured to perform a task of a plurality of tasks for the agent, wherein the larger a number of agents that perform the task is, the greater a possibility that a target for the task will be achieved increases;

• the control system comprises:
• a plurality of control apparatuses, each of the plurality of control apparatuses being configured to control a respective one of the plurality of agents, and each of the plurality of control apparatuses comprises:
• hardware, including at least one memory configured to store a computer program and at least one processor configured to execute the computer program;
• a request response processing unit, implemented by the hardware, configured to calculate, based on observation information about the agent controlled by that control apparatus, at least one other agent, and the task, a request parameter as to whether or not to request help, and a response parameter as to whether or not to respond to a request for help from the at least one other agent;
• an importance processing unit, implemented by the hardware, configured to perform processing for calculating a respective importance of each of the tasks of the plurality of tasks for the agent, based on at least the request parameter of the at least one other agent and the response parameter of the agent and a target value of the importance of the task corresponding to the observation information, the respective importance of the task corresponding to the observation information for the agent. the target value of the importance being output from each of a plurality of policies by inputting the observation information into the policy, the plurality of the policies are learned for the plurality of agents, respectively;
• a task selection unit, implemented by the hardware, configured to select the task from the plurality of tasks for the agent to be performed by the agent according to the importance;
• and
• a task execution unit, implemented by the hardware, configured to control the agent by moving the agent to a position of the selected task and controlling an arm of the agent to perform the selected task in accordance with the importance, wherein
• when the response parameter indicates not responding to the request from the at least one other agent, the importance processing unit calculates the importance of the task corresponding to the observation information on the agent such that the importance of the task being executed by the agent asymptotically approaches the target value of the importance, and
• when the response parameter indicates responding to the request from the at least one other agent and the request parameter of the at least one other agent indicates requesting help, the importance processing unit calculates the importance of the task being executed by the at least one other agent corresponding to the observation information for the agent, based on a difference between the importance of the task being executed by the at least one other agent and the importance of the task for the agent, such that the importance of the task being executed by the at least one other agent is increased.

A control method for controlling an agent configured to perform a task of a plurality of tasks for the agent, wherein the larger a number of a plurality of agents that perform the task is, the greater a possibility that a target for the task will be achieved increases;

• the control method comprises:
• calculating, based on observation information about the agent, at least one other agent, and the task, a request parameter as to whether or not to request help, and a response parameter as to whether or not to respond to a request for help from the at least one other agent;
• performing processing for calculating a respective importance of each of the tasks of the plurality of tasks for the agent, based on at least the request parameter of the at least one other agent and the response parameter of the agent and a target value of the importance of the task corresponding to the observation information. the respective importance of the task corresponding to the observation information for the agent. the target value of the importance being output from each of a plurality of policies by inputting the observation information into the policy, the plurality of the policies are learned for the plurality of agents, respectively;
• selecting the task from the plurality of tasks for the agent to be performed by the agent according to the importance;
• controlling the agent by moving the agent to a position of the selected task and controlling an arm of the agent to perform the selected task in accordance with the importance;
• and
• responsive to determining that the response parameter indicates not responding to the request from the at least one other agent, calculating the importance of the task corresponding to the observation information on the agent such that the importance of the task being executed by the agent asymptotically approaches the target value of the importance.