next up previous contents index
Next: A Taxonomy for Artificial Up: The Goal: Evolution of Previous: Distributed Intelligence - Related   Contents   Index

Multicellular Programming and Swarm-Programming

Multicellular Programming (MP) is the combination of the OOOP paradigm with genetic programming as realized in the current object-oriented ontogenetic programming system (OOOPS). It has this name because the different program modules interact with each other like the cells in a multicellular creature. Most cells have a definite position in the creature and cannot move freely. But they communicate with each other by producing diffusible substances. The production of these substances is based on and is the basis for gene regulation. In multicellular programming, the OOOP units analogously have static positions in a grid which they can only leave by dying. A new unit then can take that place only by division of an adjacent unit. They also interact on the basis of gene regulation which controls the production of diffusible messages.

Swarm-Programming (SP) is an extension and generalization of multicellular programming which allows to use the advantages of OOOP and GP also for mobile entities and very elaborate systems such as many multiagent systems. The name is written with a hyphen to distinguish it from other swarm approaches [Hiebeler, 1994,Evans, 2000]. The main differences between multicellular programming and swarm-programming are the following:

  1. The swarm units are allowed to be mobile.
  2. They take arbitrary positions in the environment. There is no grid any more.
  3. While the multicellular units usually are as small as possible, the swarm units tend to get much bigger and very complex.
  4. The swarm units' program often has to consist of much more than the OOOP part which controls behaviour and communication. Like it is not possible and makes no sense to evolve a word processor, it also makes no sense to try breeding OOOP units which can control all aspects of a real robot in a multiagent system. Such a robot has to solve so many complex subproblems for which already good solutions exist (like analyzing visual information or controlling motions) that OOOP only takes on the task of organizing the interaction between the swarm units (i.e. in this case between the robots). This means that the programs bred with swarm-programming only control the possibilities provided by the specialized subsolutions that can be developed with other techniques. The function set of the GP algorithm then for example includes functions of the communication or visual or motor control modules of the robot program. The interaction with the environment only happens indirectly through these specialized modules.
  5. Swarm-programming can include other forms of communication than the diffusible messages.
  6. Swarm-programming usually includes a much more complex environment model4 than multicellular programming.
Both multicellular programming and swarm-programming are methods for breeding intelligent distributed problem solutions. They are new and promising approaches to the hopefully soon more quickly growing field of Evolution of Distributed Intelligence (EDI).

next up previous contents index
Next: A Taxonomy for Artificial Up: The Goal: Evolution of Previous: Distributed Intelligence - Related   Contents   Index
© 2002 Peter Schmutter (