Pose-Based Tactile Servoing: Controlled Soft Touch Using Deep Learning

This article describes a new way of controlling robots using soft tactile sensors: pose-based servo (PBTS) control. The basic idea is to embed perception model for estimating the sensor pose within control loop that applied local object features, such as edges and surfaces. PBTS implemented with soft, curved optical [the Bristol Robotics Laboratory (BRL) TacTip] convolutional neural network trained be insensitive shear. As consequence, robust accurate controlled motion over various complex 3D objects attained.