Programmable Self-Organizing Cellular Automata
Loading...
Date
2022-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Cellular Automata is a paradigm for understanding complex emergent behavior in the context of simple directives. A multitude of different rulesets exist that define different progressions of this behavior and it is of interest to be able to select a specific ruleset based on some desirable end-condition of the system. This might be useful for simulating biological systems, in which one can experiment with the initial conditions to derive artificial genotype-phenotype pairings, or in robotics where self-reconfigurable swarms show promise for mapping and search-and-rescue. In this work, we introduce two means of accomplishing this goal: (1) by representing Cellular Automata graphically and taking advantage of ubiquitous graph search algorithms; and (2) by modifying a recently introduced model of morphogenetic self-organization called Neural Cellular Automata. In the former case, we define Vacancy Mapping and demonstrate its performance with Greedy Best-First Search. We find this approach to be generally satisfactory for connected reconfiguration and discuss the limitations of greedy algorithms in this context. In the case of (2), we discuss an isotropic modification to Neural Cellular Automata developed in collaboration with the model's original authors and define Structured Seeds as a way to manipulate the out-of-training behavior of the model.