Algodoo

(zoom out to see the whole scene)
This agent has 5 inputs, distance scalars, and 2 outputs, thrusters. It works for the most part, but it can get stuck against the walls. I figured out how to make local neurons, that is, not needing global variables

The first layer of the net, the Occipital Lobe, with 4 nodes, is the raw input gathered from the agent.
The second layer, The Visual Cortex with 3 nodes, represents either a kind of description of motion, the central node representing forward-backward movement, the side nodes representing rotation, or a slightly abstracted view of the environment.
The final layer, the Motor Cortex, with 2 nodes, is the output to the agent's thrusters.

The neural net has an odd beauty to it, I've noticed. Replaced the globalvars one with a localvars one, so now the brains of the agent can be toyed with.
WASD can hijack the neural net and force the agent to move your way instead of its own

hekk yea mind control wooooo