A P P L I C A T I O N S   O F   M A S

The characteristics of multi-agents systems make them appropriate to deal with applications that are modular, decentralised and changeable. Solving such industrial problems with DAI can yield solutions that are more adaptable than solutions proposed by other technologies.

Modularity

Agents have their own set of variables which interact with their environment. If our initial problems can be decomposed into different sub-modules which interact together, then modules can be assigned to different agents. Agents deal with their individual problem and communicate with each other to make the final global solution.

Decentralisation

Agents are autonomous systems; they don’t need an external control managing them. This is particularly useful in applications which can be decomposed into stand-alone processes (high level of independence between processes) because each agent performs a different task and doesn’t have to be continuously controlled.

Take for example any car manufacturer gathering pieces of a car from companies in order to minimise the cost of production. An agent-based architecture can be used to model the problem, where each agent is in charge of obtaining a specific part of the car to organise and optimise the construction. Each agent is only concerned about the piece it needs to get and not about all the pieces of the car.

Changeability

If the two previous characteristics of multiagent systems are combined, we can also use agents in problems which introduce frequent changes:
• Modularity permits modifications to the system one piece at a time.
• Decentralisation minimises the impact of a change of a particular module on the overall system (behavior of other modules).

From an industrial perspective, the ability to change quickly, efficiently and without any side effect is important. This permits a lot of variations in a manufactured product and thus a better consideration of the customers’ requirements.


Source: Multi agent systems: A Modern Approach to Distributed Artificial Intelligence, Chapter 9, Pages 377-409.