A continuous Cellular Automaton where each cell holds a vector of states (channels). Update rules are defined by convolution filters and non-linear activation functions, analogous to a localized neural network.
This model is trained (or heuristically tuned) to grow into a stable organism. The defining characteristic of an NCA is its homoeostasis and self-repair. Click and drag on the canvas to "damage" the organism (setting its cells to 0), and observe how the remaining cells coordinate to regrow the missing structure.
The same structural principle that underlies my own 'Sprouting' (芽吹き) memory system: perturbation, repair, and growth.