The coordination of actions is a set of supplementary activities which need to be carried out in a Multi-Agent environment, and which a single agent pursuing the same goals could not accomplish. It is the articulation of individual actions accomplished by each of the agents in such a way that the task being carried out ends up being a coherent and high performance operation. The main aim of these activities is to manage resources provided to agents, so as to improve the overall performance of the environment in which they are situated.

Computing procedures also have to synchronise their actions when accessing common resources. This is done to ensure that the state of the system remains coherent. There are four major reasons as to why coordination is required in a Multi-Agent System.

• In most multi-agent systems, the actions of agents are dependent on the actions of other agents. Thus, it becomes vital for an agent to interact with other agents in order to gain information and results that it may require. For example, when building a house, the building agent will need to be informed of how much land it has. This information will be provided by the measuring agent, who has measured the dimensions of the land, so it can be seen that the latter action is dependent on the first, requiring some form of coordination.


• In multi-agent systems, some shared resources may be limited in quantity. This makes it very important for agents to observe the actions of other agents and hence be aware of the lack of sufficient resources. Scarce resources need to be allocated in such a way so as to optimise efficiency of all agents. For example, let's suppose that two agents, agent A and agent B make perfumes and they require a scarce resource, say a vapouriser. If there is a total quantity of 100g of vapouriser and Agent A needs 60g and Agent B needs 50g then the 100g will need to be allocated in a fair way so that both A and B can make some reasonable quantity of their respective perfumes.


• Optimising cost by collaborating actions, through coordination: This involves avoiding repetitive actions and reducing redundancy. For example, we have an agent moving heavy objects around. By combining two separate tasks into one task, the agent can avoid the redundancy of moving the 2 objects seperately to the same place.

Now that we have outlined the reasons as to why coordination is a vital aspect of multi-agent systems, the question which may be proposed is: How do we coordinate the actions of agents?

Although the methods used to coordinate actions may differ from environment to environment, all mobile agents will need to coordinate with respect to factors such as space, distances, and speed. This is required to ensure that the movements of agents are coordinated, to avoid collisions.

Mutual Dependence between actions is one of the key points that needs to be observed when outlining a coordination plan. In other words, any form of coordination that is being proposed must be verified by all other agents that rely on it. For example, suppose we have ship agents A,B, and C. If ship A and B coordinate their actions of movement, then the plan needs to be verified with ship C to make sure that the other 2 ships do not collide with it.

Hence we can mention the following forms of coordination, that could be adopted in a multi-agent system, and are the most commonly used techniques.

• Coordination by synchronization
  
  This is the method of describing the full sequence of actions that is required to create a synchronised execution.


• Coordination by planning
  
  This technique involves the breakdown of an action into two phases. The first phase involves selecting the set of actions that needs to be carried out, and outlining a set of plans. The second phase is a a selection of one of the plans that was devised in the first phase. Coordination by planning is a general technique that can be further broken down into three distinct types. These have been outlined below :

  • Centralised Planning for multiple agents
    This involves one agent being responsible for the planning and synchronization of all actions.
  • Centralised coordination for partial plans
    This type of coordination planning involves each agent drawing up its own partial plan and then sending it to one particular agent that synthesises all the partial plans into a single overall coherent plan.
  • Distributed Planning
    This type of planning involves individual planning and coordination. There is no central agent that takes plans or coordinates the differenct actions carried out by various other agents.
  
  
• Reactive Coordination
  This takes place when there is no formal planning that takes place, coordination is done based on a particular agent's perception of the environment. For example, in a colony of ants, the ants may realize that they have sufficient storage of food, however, the water reservoir is low. Thus, the ants may react by coordinating and collaborating together to assign more ants to collect water.