Supervised learning for haptics texture classification using fourier analysis

Abstract

Human sense of touch is used to explore the environment that surrounds us, and to identify and learn about objects through the surface properties. In the design of a robotic system that is able to analyze and identify textures, it is essential to understand the perceptual factors of the human sense of touch, which presents a significant challenge in control, sensing and learning. However, recent developments in haptic sensing have made it possible to explore surface textures and classify them through a learning algorithm. This thesis investigates the use of Haptic feedback as an approach to improve and classify surface textures by a robotic system. A review of haptic interactions indicated that haptic information provided by the sense of touch, are used successfully to convey important data of the surface texture properties. Hapti c feedback, expressed through the kinesthetic measurements of the surface waveform that arises when prescribing a predefined motion over the surface texture, was collected from four cardboard samples wit h different surface properties. The motion trajectory traced by the spherical probe shows some intrinsic properties that facilitate the data extraction and that reproduce the way humans identify the texture of a surface. Features were extracted from this data through frequency spectrum by Fourier analysis and used for training and classification by a supervised k-NN machine learning algorithm.