Towards Applied Tactile Sensing: Methods and Algorithms for Real-World Robotic Applications with Robot-Centric Touch Perception

This dissertation explores methodologies for robotic control and perception which leverage artificial sense of touch, provided by large-area tactile sensors. The whole thesis is developed around the concept of robot-centric perceptual architecture. This means having a sensing modality completely distributed along the robot body and capable to acquire rich information about the environment such as its stiffness or geometry, without any external sensors placed in the surrounding region. The corpus is divided into three main chapters which follow the progression of the author’s research path. The first chapter aims to demonstrate how large-area tactile sensors allow a robotic manipulator to operate in unmodeled environments. The developed algorithms are conceived by taking into account industrial scenarios in which robots with a closed control architecture are usually employed. The second chapter expands the concept of unpredictable environment interaction to human elements. It shows how distributed tactile sensing can be used as non-verbal communication interface for Physical Human- Robot Interaction (pHRI). The third main part of the dissertation focuses on a major practical issue of tactile sensors, which nowadays limits their application in real-world scenarios. The lack of a standard data representation and the differences in tactile sensors technologies and implementations are add synthetic tactile data generation via state-of-the-art data-driven techniques.