Discovering Articulations by Touch: A Human Study for Robotics Applications

Abstract

Getting robots to manipulate arbitrary objects in unstructured, human-centric environments is an open problem. Typical assumptions for engineering-based solutions to this problem are that the object can be recognized, that it is nearly rigid, and that its pose can be estimated. Even under these stringent assumptions, autonomous performance of such tasks has remained primitive over decades of research and development. The reasons for the slow pace of advancement are generally unknown, though careful engineering of robotic systems to perform such tasks continues. This paper attempts to investigate why by examining human manipulation performance on a task that appears heavily dependent upon tactile feedback, a known deficiency in current robotic manipulation technology. Current contact sensors are noisy, bulky (preventing precisely localized contact sensing), wear out, or—commonly—some combination of the three. This paper presents the result of a human study to understand the importance of haptic kinesthetic feedback to discover articulations upon encountering a novel mechanism. The findings of the study indicate a decrease in performance of a complex manipulation task when haptic feedback—as the primary sensory input—is desensitized.