Object recognition using tactile array sensors

In this dissertation, we explore the use of tactile force sensors to understand properties of the surfaces of objects they are pressed against. We develop a model of such sensors as imaging devices, which facilitates the use of techniques from computer vision and image processing with the "tactile images" they provide. The goal is object recognition, and three approaches are presented for distinguishing amongst objects from a previously-encountered set: The first approach is entirely geometric in nature and borrows ideas from mobile robotics. Object surfaces are modeled as occupancy grid maps, and recognition is posed as a problem of localizing the sensor within one of these maps. The second approach applies techniques from computer vision to characterize the tactile appearance of objects. Finally, the third approach combines information from the previous two to describe the spatially-varying appearance of objects' surfaces. These methods are evaluated in experiments using a physical tactile force sensing system and in a variety of simulations based on our imaging model, and they all exhibit strong performance in their respective domains.