The role of tactile sensing in robot manipulation

Abstract

Robots must sense their environment and be sensitive to the changes taking place if they are to move beyond performing a limited range of tasks in the structured surroundings of a factory, to performing a variety of tasks in an unstructured and unknown environment. This thesis addresses a key requirement for such robots, namely, the capability of manipulation based on touch sensing. Tactile array sensors provide information about the distribution of force and other parameters within the area of contact. To manipulate objects using information from these sensors, it is necessary (1) to understand these sensors, i.e., have a model of them, and then (2) to develop robot control strategies to do a given task. The first contribution of this thesis is the development of models of contact between a tactile array sensor and objects of different shapes, specifically a knife edge, a rectangular bar, and a cylinder. Many assumptions are made in elasticity theory to obtain closed-form solutions for such problems. Some of these assumptions are not appropriate for studying tactile sensors. This is most notably true of the half-space assumption in which the sensor is modeled as a semi-infinite elastic medium bounded by a plane. However, in the absence of this assumption, closed-form expressions cannot be obtained for many cases. The finite element method is presented as a general and convenient method to study such problems. Using this method, the sensor is modeled as a thin layer and sub-surface stress distributions are obtained for contact with different objects. The second contribution of the thesis is the development of a new approach to robot motion control during manipulation. The approach presented here is called tactile servo and is based on information from a tactile array sensor. Analogous to image-based visual servo, tactile servo is based on the observation that the progress of a manipulation task can be characterized by the tactile images obtained from sensors mounted on the fingertips of a robot hand. The sub-surface stress distributions obtained using the finite element method are analyzed to identify various features that are predictive of the applied load given the shape of the contacting object. Amongst these features, the moments of the stress distributions are found to vary in a regular way with the applied load. Hence, the zeroth, first, and second order moments of the stress distributions are selected as the tactile features for use in tactile servo algorithms. A motion control algorithm based on tactile servo is developed for the task of rolling a pin on a planar surface using a planar robot finger equipped with a tactile sensor array. This algorithm is also implemented experimentally to show that information from a tactile array sensor can be used in a simple, direct and effective way to control manipulation tasks.