Neural Cellular Automata (NCAs) have been proven effec- tive in simulating morphogenetic processes, the continuous construction of complex structures from very few starting cells. Recent developments in NCAs lie in the 2D domain, namely reconstructing target images from a single pixel or infinitely growing 2D textures. In this work, we propose an extension of NCAs to 3D, utilizing 3D convolutions in the proposed neural network architecture. Minecraft is se- lected as the environment for our automaton since it al- lows the generation of both static structures and moving ma- chines. We show that despite their simplicity, NCAs are ca- pable of growing complex entities such as castles, apartment blocks, and trees, some of which are composed of over 3,000 blocks. Additionally, when trained for regeneration, the sys- tem is able to regrow parts of simple functional machines, significantly expanding the capabilities of simulated morpho- genetic systems.
2103.08737-2