(Emergent Morphology)
1 Introduction
In this paper I present an animation system developed for exploring the emergence of functional
anatomy and motion behavior. Artificial life research projects typically apply genetic
algorithms (GA's) (Holland, 75) as a means of achieving emergent behavior in artificial
organisms (Langton, 89). While many types of emergent phenomena have been modeled,
morphology still remains an area with room to explore. The techniques and discoveries
I briefly present here are derived from research in character animation in which
motions are generated by forces internal to the character, using physical modeling techniques
such as forward dynamics (Badler et al., 91). GA's have been used to optimize motor
control schemes for locomotion and other behaviors in physically based characters
(Ngo and Marks, 93). This paper describes a variation on these techniques - with an added
emphasis on emergent morphology driven by fitness pressures for locomotion and head
motion constraints during evolution.
The stars of the show are biomorphs modeled as 3D articulated stick figures
(called animats here) existing within a simple physical environment. I have chosen not
to put flesh on the skeletons because my focus is primarily on motion and gross morphology,
and 3D line-drawings serve well for this purpose. Anatomy and motion attributes are represented
by a few dozen parameters in each animat, which evolve through the use of a GA.
Relative fitness among animats is defined as the ability to travel longer distances from a
common starting point. Secondary fitness terms are used which add pressures for holding
the head high, minimizing head movement, and minimizing head contact with the ground.
