Consider an arbitrary grid of cells with random identities. In Toddler these cells are always be moving, as they draw from the real motion of a toddler moving about a room. Here, for the purposes of illustration, we draw these cells in an abstract data space.

Inspired by models of cell-differentiation in embryology, these cells constantly send out signals to each other, trying to assign cells to nodes in the underlying motion, seeking to form patches of nodes with the similar identities.

Global structure emerges from local rules—over time as a cell becomes the child’s finger, a nearby cell becomes the wrist. We can bias our collection of cells to take on particular forms by pinning the identities of certain cells or modulating the force of the signal transmitted by each cell.