Topos: Spiking neural networks for temporal pattern recognition in complex real sounds

This article depicts the approach used to build the Topos application, a simulation of two-wheel robots able to discern real complex sounds. Topos is framed in the nouvelle concept of subsymbolic artificial intelligence, applied to the field of evolutionary robotics. This paper focuses on the simulation of biologically inspired artificial cochleas and spiking neural networks, in order to model the embodied control system of the robots. The method chosen to find the most appropriate parameters that determine robots’ behaviour is evolutionary computation techniques, with the aim of avoiding any human intervention in this task. As an example of a real application of this technique, experiments were performed to study the ability of the robots to distinguish sounds composed of parts of real canary songs and to navigate to the recognised signal. Results obtained confirm the validity of the approach.