Artificial morphogenesis via 3D neural cellular automata

Morphogenesis is the biological process by which an organism grows and takes its form. It is responsible for the growth and shape of all living things and has historically been studied from the perspective of developmental biology. Accordingly, experiments have routinely been performed using real organisms and their cells. However, with continued advancements in computational capabilities, researchers have been able to simulate physical and biological systems with greater fidelity and accuracy. A recent example is the development of two-dimensional neural cellular automata, which have the ability to artificially model morphogenesis. Cellular automata, the antecedent to neural cellular automata, are a type of computational model uniquely suited to model physical systems due to their inherit discretization of space and time, locally rule-based dynamics, and embarrassingly parallel execution. Their ability to exhibit rich and sophisticated results from seemingly trivial implementations has lead to their continued use to study emergent properties found in nature. This thesis expands upon previous works by presenting a novel three-dimensional neural cellular automata which illustrates self-organizing, regenerative, and isotropic properties. Understanding the complexities of morphogenesis could provide beneficial insights for the development of regenerative medicines, self-organizing robots, and other bioengineering endeavors.