UC Berkeley

Accurate dynamic object manipulation in a robotic hand remains a difficult task, especially when frictional slip is involved. Prior solutions involve extensive data collection to train complex models to control the hand that do not necessarily generalize to other slip circumstances. Our approach focuses on direct slip sensing using a tactile sensor with a capacitive array, coupled with a programmable system on a chip, capable of mode switching and sampling rate adjustment. We characterize the sensor's capacity to sense slip features at higher speeds and introduce a novel methodology for estimating motions. Low-level sensor reprogramming that couples multiple taxels improves slip avoidance and reaction time during rapid slip onset events. The technology also tracks dominant surface vibration frequencies resulting from stick-slip cycles, estimating speed and acceleration of smooth flat surfaces. Using a parallel-jaw robotic gripper, we demonstrate dynamic repositioning of objects lacking trackable surface features within the hand. The goal of this investigation is to support faster reasoning and reflexes for dynamic dexterous robots that experience directional in-hand slip.