Agent communication and co-operation

Agents may communicate with other agents and non-agent processes directly via asynchronous message communication. This may be addressed communication with the identity of the receiver and sender attached to the message, or indirect communication using a publish/subscribe server. Additionally, agents might communicate indirectly by querying and updating an external shared memory, such as a data base or Linda tuple store[1]. They may even communicate by making certain changes to the environment state, e.g. the analogue of the pheromone trails left by ants.

A robotic agent RA queries its DBS to determine which device action A to use next for a particular task, T. If a group of agents are co-operating on T, this might be an action of a device controlled by another agent RA. RA may need to explicitly request that RA initiate the action, or RA may automatically do the action when it determines what needs to be done next for shared task T. Either way, A will be sent to the simulation, or to a ROS[6] or MQTT[5] node that has direct non-agent control of the robotic device in question. A is being done to try to bring about a change in the common external environment of RA and RA that will progress the task T.