This baby should be ready for play later this week. I've got the menu system in now. The main menu allows you to choose between Learning and Evaluation modes. There are 4 memory slots from which the user can choose. This allows the user to compare different agents in the evaluation mode. The training mode graphics look a little better now with drop shadows added and grass that grows. The training section still uses the tank-like graphics but the evaluation mode uses an actual lawn mower graphic.
The saving of the networks happens automatically in the learning controller. When an agent's fitness score exceeds the previous dominant agent's fitness score then its neural network is saved in the chosen memory slot.
I have three more things to finish. Complete the evaluation controller (allows the user to move a rain cloud around to grow grass and test a trained agent's cutting skills), add sound effects, and refactor / clean up the code.