(published in: From Animals to Animats (page 484) 1996, MIT Press)
Sexual Swimmers
Emergent Morphology and Locomotion Without a Fitness Function
Jeffrey Ventrella
Abstract
A virtual ecosystem is described which demonstrates emergent morphology
and locomotion among a population of 2D figures in a virtual pond. A
physics model enables a large variety of articulated figures to potentially
propel themselves through simulated water. Swimmers with better skills
at locomotion and turning are able to eat more food to gain energy and
mate with other swimmers, thereby supplying more "fit" genetic building
blocks to future generations. There is no fitness function used in this
genetic algorithm scheme: fitness equals reproduction. Sexual selection
is also modeled: genetically inherited preferences for mate color affect
the distribution of phenotypes across the pond: swimmers of differing
colors infrequently mate with each other, often breaking the population
into distinct coloration groups. To witness the emergence of behavior
on local scales (swimming styles), intermediate scales (family dramas),
and global scales (population dynamics), this simulation can be viewed
with an interactive "microscope" for panning and zooming. This work
demonstrates how local behavior can affect (and be affected by) global
behavior, and how these different levels of behavior evolve together.
