Friday, May 10, 2013

Evolution as a Tool to Design Self-Organizing Systems

Self-organizing Systems exhibit numerous advantages such as robustness, adaptivity and scalability, and thus provide a solution for the increasing complexity we face within technical systems. While they are attractive solutions, due to their nature, designing self-organizing systems is not a straightforward task. Artificial evolution has been proposed as a possible way to build self-organizing systems, but there are still many open questions on how an engineer should apply this method for this purpose.
Evolutionary design process
We propose a system architecture for evolving self-organizing systems that marks the major cornerstones and decisions the designer has to face:


In particular, the following aspects need to be considered:

Simulation setup:
  • Accuracy/granularity of the simulation?
  • Physical capabilities of the agent + environment
  • How many agents, homogeneous/heterogeneous configuration?
Interaction interface:
  • How should the agent interact with the environment/other agents
  • Number/type of sensors
  • Ability to change the environment (enable stygmergy)
Evolvable decision unit
  • Must be evolvable
  • Smooth search space, not too large
  • Genotype-to-phenotype mapping
Search algorithm
  • There exists literally a zoo on metaheuristic optimization algorihtms (Cuckoo search, Honeybee, Frog leap, Firefly, ...)
  • Ability to find global optimum
  • Number of tweaking parameters?
Objective function
  • Should contribute to a smooth search space
  • Avoid modeling the solution instead of the problem
  • Mapping of multiple objectives (or weighted sum?)
Framework for Evolutionary Design (FREVO)
  • Is a way to implement an evolutionary design task for multi-agent system
  • Needs a simulation of the problem
  • Interface for sensor/actuator connections to the agents
  • Feedback from a simulation run -> objective function
  • Written in Java, runs on multiple platforms including Linux, Mac OS, Windows
  • FREVO is available as open soure at http://frevo.sourceforge.net/
  • FREVO introduction video including installation, setting up a simulation and running it (length 6 minutes)

Literature