I wonder how much damage(physical damage) comes into play with evolving robust systems and creating interesting variations in nature. Using the robotic gait example, in the real world organisms have a high enough chance of losing limbs that having the neural circuitry to survive in the absence of one or more limbs is advantageous. Maybe these simulations could increase their ability to produce robust networks by modeling accidental damage(from tiny to catastrophic) to random parts of networks and selecting those that survive in spite of such damage.