A Control System for Multi-modal Stereognosis

This work describes the architecture of an integrated multi-modal sensory (vision and touch) computational system. We propose to use an approach based on robotics and control theory that is motivated by biology and developmental psychology, in order to integrate the haptic and visual information processing. A robotic agent has to deal with its dynamic environment learning how to construct a map and how to keep this map consistent with a current perception of the world. We consider two policies for the system control. The first is developed using a feedforward strategy and the second we have derived using Q-learning. We show the results carried out in the simulation environment. Basically, as a practical result, the system is able to foveate on regions of interest of its environment, and to change its focus of attention from one region to another. Also, an efficient feature extraction takes place allowing object categorization. This work developed in simulation is the basis for the impl..