| An array of force or displacement sensors packaged in a robotic finger can extract local object shape, size, and orientation information necessary to perform dextrous rolling manipulations. These tactile array sensors can also provide information about the local finger/object contact to aid in grasp stability analysis. In this thesis, we propose both a contact mechanics-based and a geometry-based method for estimating an unknown object's local shape from data from a tactile array sensor. Both methods are very simple and fast and are applicable to flat, spherical, cylindrical or ellipsoidal fingers and objects. A capacitive array sensor was designed to best complement these methods. Sensor material is shown to significantly affect both sensor sensitivity and dynamic response. The sensor was modeled using solid mechanics and approximately measures surface stress. Experiments using the sensor and the geometric method were able to estimate an unknown object's radii of curvature to within about twice the sensor spacing. Accuracy can be significantly improved with simple fitting algorithms, although at the cost of computation speed. Simple rolling experiments confirmed that our geometric method can be used for real-time, sensor-driven rolling manipulations of unknown objects. Various types of tactile array sensors were examined, and surface displacement array sensors were found to be best-suited for estimating local object geometry. Finally, experiments were conducted to investigate the effects of soft, deformable fingertips on rolling kinematics. For fingertip deflections of 10%, rolling trajectories were shown to deviate up to 5% from rigid body predictions depending on the fingertip material. These deviations can be minimized with an appropriate choice of material in finger designs. |
