Learning Slip with a Patterned Capacitive Tactile Sensor

Yuri Gloumakov, Tae Myung Huh

Keywords: Tactile Sensing, In-Hand Manipulation

TL;DR: Detecting slip speed of a flat object sliding across a tactile sensor

Abstract: The task of dynamically manipulating objects within a robotic hand presents ongoing challenges. In particular, friction and slip often dictate task success yet remain difficult to measure directly, quickly, and accurately; this includes both the detection of slip events and slip speed. Complex solutions exist that involve training a control policy using neural networks, with image-based sensors or external cameras, or when contact geometry can be inferred. Using only a capacitive sensor with a `nib`-patterned structure, we attempt to demonstrate the sensor’s ability to detect slip speed during uninterrupted contact where geometry cannot be inferred, while benefitting from faster sensing, cheaper construction, and smaller profile. We hope that by collecting vibration amplitude and frequency and applying supervised learning techniques to directly measure slip speed we can guide an implementation of manipulation controls without a priori assumptions about object properties, such as friction or geometry.