AgentCoordination

context: "In an environment with limited resources, agents must coordinate their activities with each other to further their own interests or satisfy group goals. The actions of multiple agents need to be coordinated because they are dependencies between agents' actions, there is a need to meet global constraints, and no one agent has sufficient competence, resources or information to achieve system goals." (Huhns & Stephens, in Multiagent Systems book)

Therefore the key aspects are:

• the activities of the agents should solve all the components of the global problem
• the interactions between agents should be coherent and integrate in the global solution

Good coordination should:

• provide timely information and help to the agents
• avoid redundancies between agents actions
• make sure that the agents optimize their use of resources for their tasks instead of communication and coordination.

1. Task allocation: who should do what, how... Tasks could be allocated a priori using predetermined AgentOrganization, or dynamically using ContractNet.
2. Planning: how to find the right sequencing of subgoals and subtasks allocated to the different agents in order to achieve the global goal? Plans could be determined in a centralized way (using for instance STRIPS or NOAH [Sacedotti]) and then sent to the various agents [Corkill 79]. Or plans can be calculated locally, and the agents would then send their partial plans to each other [Durfee 87]
3. AgentNegotiation

protocols:

• ContractNet for task allocation
• AgentNegotiation protocols