Evolving Spiking Neural Networks to Control Animats for Temporal Pattern Recognition and Foraging

We evolved spiking neural networks (SNNs) to control animats in a task requiring temporal pattern recognition and foraging in a 2D environment with two types of objects emitting patterns: a target and a distractor. The target emits a specific temporal pattern composed of two components, while the distractor emits random patterns that are all the other combinations of these two components. The fitness function rewarded finding targets and avoiding distractors. We show that the evolved animats are robust to changes of the number of objects in the environment, strength of the actuators, duration of signals, intervals between signals in the pattern and between patterns. Our long term goal is to understand the mechanisms governing the neural networks that accomplish simple but not trivial computational tasks inspired by minimally cognitive abilities of animals, such as phonotaxis.