Tactile perception of cognitive robots

Compared to other modalities like vision, tactile sensing has been so far neglected by robotic researchers. At the time of writing, tactile sensing devices that can just approach the performance of the human sense of touch seem out of reach. Despite this fact the use and exploitation of available sensors should not be disregarded. Tactile sensing is indispensable for in-hand manipulation and can reveal object properties that cannot be acquired by optical sensors. The aim of this dissertation is to exploit tactile sensing in robotic setups. We begin by providing a summary of interesting findings from the neurophysiological community. After the description of preliminary work on the experimental setup, the next chapters deal with the artificial sense of touch. Tactile object recognition is one of the most matured areas of applied tactile sensing in robotics. Two very different and novel approaches in this domain are presented. The first one introduces the idea of entropy in tactile stimuli. In the second approach, an artificial robot hand equipped with tactile sensors was used to probe previously unseen objects. A well known probabilistic method was employed to enhance the gathered information. The acquired 3-D point clouds open up many new possibilities to improve the perception of the robotic system, for example through sensor fusion with 3-D vision systems. Object recognition is a welcome by-product of this method. We continue in this thesis focusing on optimal grasp force determination through slip detection with a static tactile sensor. Detection of incipient slippage is a very important aspect of tactile sensing as it allows for the adjustment of the grasp force without any prior knowledge about the weight and friction of the object. Finally we use touch to deal with one of the great challenges in robotics: the handling of deformable material.