Slip detection on natural objects with a biomimetic tactile sensor

Abstract

Slip detection enables robotic hands to perform complex manipulation tasks by predicting when a held object is about to be dropped. Here we use a support vector machine classifier to detect slip with a biomimetic optical tactile sensor: the TacTip. Previously, this method has been shown to be effective on various artificial stimuli such as flat or curved surfaces. Here, we investigate whether this method generalises to novel, everyday objects. Five different objects are tested which vary in shape, weight, compliance and texture as well as being common objects that one might encounter day-to-day. Success of up to 90% is achieved which demonstrates the classifier’s ability to generalise to a variety of previously unseen, natural objects.