Increasing the complexity of solutions produced by an evolutionary developmental system.

Evolutionary computation and neuroevolution seek to create systems of ever increasing sophistication, such that the digitally evolved forms reflect the variety, diversity, and complexity seen within nature in living organisms. In general, most evolutionary computation and neuroevolution techniques do so by encoding the final form without any type of development. This is in contrast to nature, where most complex organisms go through a developmental period. Here we focus on an evolving digital tissues that develop from a single cell and unfold into a complex body plan. It quickly became evident that evolving developing forms is quite challenging. We compare four different techniques that have successfully been employed within evolutionary computation to evolve complex forms and behavior: scaffolding (i.e., gradually increasing the difficulty of the task rewarded by the environment over evolutionary time), stepping stones (i.e., rewarding easier tasks within an environment that can co-opted for the performance of more complex tasks), and island models (i.e., rewarding different fitness functions within different subpopulations with migration). We show the effect of these methods on the evolution of complex forms that develop from a single cell, the rate of adaptation, and different dimensions of robustness and variation among solutions.