[ Log In | Sign Up | Lost password? ]

Agents & Multi-Agent Systems


An agent is small program object with underlying constraints C and a utility function u - describing more or less numerically its satisfaction. The latter is done by mapping the solution state S onto a numerical value. Agents can predict the outcome of a state change by making assumptions on the utility increase in accordance with its decision. Typically this assumption is modelled by probability distributions derived for a given problem.

Multi-Agent Systems

A system with N>1 agents is called a multi-agent system, if a language model exists so that the agents can communicate with each other. Such systems can be used to solve optimisation problems, e.g. for problems with distributed constraints, where a formal approach is not tractable. Various industrial applications have successfully used multi-agent technology and there is an active research community that expands the applicability of this technology as tool continuously. Multi-agent systems are closely related to state-machines and event driven systems.


Agents can form so-called coalitions with other agents to perform tasks. Once different species of agents must group together to solve a defined task, this coalition is called a holon. In other words, without building the special holon, different agents or agent coaltions can in principle not be succesfull with their tasks, as certain skills are missing in their group.

Holons have become a popular means in the optimization of production processes as well as in the fields of scheduling and planning.

One of my primary research interests is the formal mathematical description of these kind of systems. As to my knowledge, there is no closed theory, yet, and the prediction of both stability and robustness relies essentially on simulation and test results, rather than on formal stability theorems.

Evolving multi-agent systems

Genetic evolution

A special class of multi-agent systems applies concepts of evolution. In close relation with the evolutionary ensembles, we evolve a set of agents towards increased fittness. Difficult is finding the right metrics that is applied to tackle the fittness parameter and mostly this is already the key to solve the problem at hand.

Artificial memetisms

In multi-agent systems, communication is the element that drives the problem solving. Memetisms are evolutions within the communications and the behaviour of the agents. Artifical tradition and communication structures, that can be exploited to force the system into mimmicking existing problem scenarios.