Haptic feature extraction from a biomimetic tactile sensor: Force, contact location and curvature

摘要

The BioTac® is a biomimetic tactile sensor for grip control and object characterization. It has three sensing modalities: thermal flux, microvibration and force. In this paper, we discuss feature extraction and interpretation of the force modality data. The data produced by this force sensing modality during sensor-object interaction are monotonic but non-linear. Algorithms and machine learning techniques were developed and validated for extracting the radius of curvature (ROC), point of application of force (PAF) and force vector (FV). These features have varying degrees of usefulness in extracting object properties using only cutaneous information; most robots can also provide the equivalent of proprioceptive sensing. For example, PAF and ROC is useful for extracting contact points for grasp and object shape as the finger depresses and moves along an object; magnitude of FV is useful in evaluating compliance from reaction forces when a finger is pushed into an object at a given velocity while direction is important for maintaining stable grip.