|I ran the simulation from Saturday afternoon till today and it ran well. However, the robots knocked all the food blocks away some how.|
The robots keep circling in one direction and that doesn't seem right. One of my options is to reconsider how the output neural networks work with the actions. Currently, one neuron controls whether robots move forwards and another neuron controls whether robots move left or right. This doesn't feel right. Ideally, for example, I would like to have a number of neurons control whether a robot bends or extends a leg. The first key point is that there needs to be a group of neurons that fire in a coordinated way and the second key point is that movement should require coordination of limbs.
My first goal is to make sure that the neural network is activating and adapting properly. My idea is to somehow visualize a neural network. It would be best if I didn't have to program a visualizer but I want one that integrates into the simulation. I've had a look before and I took a look again for a program I can use but there doesn't seem to be one that feels right.
Grr. I have to change the way the simulation loads values from files so that I can more easily move member variables to more convenient locations. Then I can keep going with my code to visualize neural networks.
. . .