Development of Abstract Categories in Embodied Agents

In this paper we demonstrate how a neuro-robot situated in an environment containing parallelepiped objects that vary in shape, size, and orientation can develop an ability to recognize and label the category of the objects and generalize to new objects. The analysis of the dynamical system constituted by the robot and the environment in interaction allowed us to understand how adapted agents solve the categorization problem at the level of the detailed mechanisms and at the level of the general strategy.