Yavanna
Evolving the morphology and behaviour of virtual creatures

Introduction:
Yavanna uses evolutionary algorithms and a physical-world simulator to optimize the genotype of virtual organisms like the ones above. These pictures were taken from a MPEG-movie by Karl Sims. The creatures used in Yavanna are based on this kind of virtual organisms.
These two papers by Karl Sims describe the kind of organisms used in Yavanna
[1] Evolving 3D Morphology and Behavior by Competition
[2] Evolving Virtual Creatures

Research:
Research topics using the physical-world simulator as a base could include:

• Multiple level genetic algorithms
  The genetic algorithm itself can also be evolved which will increase the speed in which the evolution reaches reasonable results. Also, two or more levels of this meta-evolutions could be used and the results compared with the single level meta-evolution.
• Social structures
  The organisms could be simulated simultaniously in a way they have to work together in order to survive. Social structures like tribes will then emerge..
• Self-Organising vocabularies [3]
  If we add a communication device like a sound emitter/receptor self organising vocabularies may emerge and more complex tasks could be possible for the population because they will be able to communicate easily.
• Morphogenesis
  The questions asked here are: Is there or are there ideal morphologies for a given kind of medium ( water, air ) in which the organisms have to survive? What is the relationship with the morphologies found in nature? How does evolution reach this morphology, is the path followed by the evolution always the same?
  If so why does evolution always reach that kind of morphology, what 's different about this ideal morphology?
• Ideal behaviour
  The same questions as in morphogenis can be posed about the behaviour or nervous system structure when the morphology of the organism is fixed. Another question could be : how does evolution adapt an organism to another environment (e.g from water to land) without being allowed to change its morphology?
• Start idependent behaviour
  By added high level receptors, neuron functions and effectors (e.g. sound input, sound recognition, sound emitter) and combining this with a (self-organized) vocabulary the organisms could be given tasks during evolution and eventually learn to complete these tasks without a fixed starting position.
• Numerous other examples can be given...

Some practical examples of the above research topics are being prepared right now :

Insect Handmade organism resembling an insect. It contains no brain... An appropriate nervous system could by evolved in a way that the organisms starts walking like real 6-limbed insect.

Simulation A handmade swimming organism. Its fitness is being determined by letting it swim for some time and measuring the length of its trajectory.
These organisms could be used to seed an initial population and continue evolution. The random-organism creator could also be used when evolution has to start from scratch.

Software:
The archive contains the evolutionary-algorithms, organism visualizer, physical-world simulator.
Uses standard GNU C++, GNU make, Mesa-OpenGL libraries ( tested under Linux, Solaris )
yavanna.tgz

References:
[1] Sims K., "Evolving 3D Morphology and Behaviour by Competition" Artificial Life IV proceeding, MIT Press 1994
[2] Sims K., "Evolving Virtual Creatures" SIGGRAPH '94 Proceedings, July 1994 p15-22
[3] Steels Luc, "Self-organising vocabularies" ARTI-lab VUB.
[4] Brooks Rodney A., "Intelligence Without Reason" Computers and Thought, IJCAI-91

Authors:
Johan Kaers Genetic algorithms
Tim Aerts Physical-World Simulator