This paper investigates the problem of artificial perception related to manipulation tasks in robotics.
The proposed approach is based on biological models of perception and sensory-motor co-ordination in humans and aims at devising anthropomorphic solutions to the problems of perception, learning and control in robotics. In particular, our approach involves the integration of different sensory modalities and the interpretation of sensory data aimed at the control of robot behaviour.
We consider as sensory modalities, in relation to manipulation tasks, vision and haptic perception, i.e. the integration of tactile proprioceptive and exteroceptive data.
The experimental part of this work is aimed at investigating some aspects of the proposed anthropomorphic model of perception in manipulation by means of anthropomorphic visual and tactile sensors on a robotic manipulator and a pantilt head, and a processing module based on neural network computational models integrated with the reinforcement learning paradigm.