Learning high-level sensors from reflexes via spiking networks in roving robots

In this paper we introduce a network of spiking neurons for navigation control. First, the robot is equipped with a system of spiking neurons able to avoid obstacles. Then, a second layer is designed with the aim of providing the robot with a target approaching system, able to direct the robot itself towards visual targets. In both cases we assume that the robot knows some a priori response to low level sensors (i.e. to contact sensors in the case of obstacles or to proximity target sensors in the case of targets) and has to learn the response to high level stimuli (i.e. distance sensors or visual input). Spike-timing-dependent-plasticity (STDP) is used to make the system able to learn high level responses.