The feature of memory could also be created within the neural network by creating tapped-delay lines for each of the inputs (extending the input vector from one to two dimensions). Therefore, the prior input from the environment doesn't disappear, but becomes part of another input to the network. This could be extended for a number of elements giving the character quite a bit of history, as well as many new intelligent abilities .
This method could also include feedback, so that the last actions could be fed back into the network providing the neurcontroller with action history. Additional internal feedback could include the health of the character as well as the simulated affective state (the character's emotional state). These mechanisms further the generation of rich and believable action selection.