Using the our existing GP library and the additions we made to it for specific traffic signal control, we ran a large number of experiments to determine the usefullness of our approach. During the course of these experiments, we were able to determine that the genetic creatures could do well at finding small adaptive control laws for moderately loaded network. The GP often came up with clever combinations of the functions that proved to be based on reasonable traffic control strategies.
One of the more interesting results of this research, was the GP's surprising ability to use the constants to its advantage. Since this representation is a combination of GP and GA approaches, how the two parts of the creature interacted in both evolution and execution was interesting. See demonstration 1 for one particular way the creature evolved its constants to fit the problem.