The Project

The creatures project aims at investigating moving and learning agents in artificial worlds. These agents are called creatures in the following. Our  investigations concern the optimization of the behavior of creatures with a local view but global tasks in multi-agent- respectively multi-creature-systems.

There are many fields of applications for such artificial worlds:

  • Computational worlds: Creatures are considered as active moving objects. Passive objects contain data. Creatures are programmed or are able to learn to solve a complex algorithmic problem.

  • Synthetic worlds: Games, genetic art, optimization of the behavior of the creatures to reach global goals, social behavior, self organization.

The world for the creatures is modeled in cellular automata (CA) because of the massively parallel nature of CA and the time consuming simulation of such multi-agent-models. The CA allows us to describe a time- and spatially discrete world that can also efficiently be supported by multiprocessor systems or special hardware.

The general goal of the project is to develop methods which allow to find the optimal local behavior of moving creatures in order to solve a given global problem.

 

Example: The Creatures' Exploration Problem

The creatures' exploration problem is a simple example of a multi-creature-system in an artificial world.

Given is a 2D-CA environment with obstacles and moving creatures. The global task of the creatures is to visit all the empty reachable cells at least once with a minimum number of generations, while each creature does not know about the global task, but has only a local view and thus a local behavior rule. One restriction for this specific problem is the exclusion principle, i.e., only one creature can be on one cell at the same time.

Possible applications are "mowing a lawn", "vacuum cleaning a room", "exploring an unknown environment" or "distribute information".

The goal now is to optimize the local behavior of the creatures to obtain a good global behavior. The questions that arise are:

  • How should the behavior of a creature be modeled?
  • Which methods can be used for optimization?
  • How many creatures are optimal?
  • How can behaviors be developed that are robust to changes of the environment?
  • Can the results for this specific problem be generalized, i.e., can we use the same optimization methods or the same modeling of behavior for other problems, too?
Three creatures trying to explore a 2D-environment with obstacles. Initial configuration (left) and after some generations (right).

Simulation Program

For the CEP (Creatures' Exploration Problem) a simulation tool is available for download. In the file CEP_Simulator.zip you find the executable JAVA program files and a users manual.

 

A A A | Drucken Print | Impressum Impressum | Sitemap Sitemap | Suche Search | Kontakt Contact | Website Analysis: More Information
zum Seitenanfangzum Seitenanfang