Multiscale Dynamic Learning in Cognitive Robotics

This paper is concerned with the dynamics of Cognitive Developmental Robotic architectures and how to produce structures that allow these types of architectures to deal with the different time scales a robot must cope with. The most important types of dynamics that occur in different time scales are defined and different mechanisms within a particular cognitive architecture, the Multilevel Darwinist Brain, are suggested to model each one of them. The paper also proposes a novel neuroevolutionary technique, called τ-NEAT, in order to capture processes based on precise temporal cues. This technique is analyzed when addressing dynamic environments in a real robotic test.