A central pattern generator (CPG) is built to control a mechanical device (
plant) inspired by the pyloric chamber of the lobster. Conductance-based mo
dels are used to construct the neurons of the CPG. The plant has an associa
ted function that measures the amount of food flowing through it per unit o
f time. We search for the best set of solutions that give a high positive f
low of food in the maximization function. The plant is symmetric and the mo
del neurons are identical to avoid any bias in the space of solutions. We f
ind that the solution is not unique and that three neurons are sufficient t
o produce positive how. We propose an effective principle for CPGs (effecti
ve on-off connectivity) and a few predictions to be corroborated in the pyl
oric system of the lobster.