Compliant artificial skins to enable robotic sensing and training by touch

Abstract

Co-robotics is interested in understanding how robots interact with other structures, such as humans, for the purpose of decision-making. Robots are currently designed to execute programmed actions to satisfy specific commands and motions. If a robot could be taught to perform an action by a guiding touch from a human, rather than specific programmed actions, a single learning approach could be broadly applied to multiple robotic platforms. In order for a robot to learn from touch, the robot must have a sense of touch. To provide the robot with this sense of touch, an artificial skin was created by applying a conductive exfoliated graphite/latex mixture to a compliant latex substrate, resulting in a highly compliant skin conformable to a variety of structures. The skin was designed with an array of strain gauges on both sides of the substrate to form a grid capable of detecting localized forces through changes in skin resistance. Model experiments were used to characterize the mechanics of these skins when placed over compliant substrates. Surface strain was characterized directly with 3D DIC, and shown to correspond quantitatively to changes in skin resistance. The ability of the skin to sense touch has been demonstrated using a Braille target and a commercial robot arm with a finger designed with the sensing skin. Using these skins, it will be possible for robots to feel and sense features of their environment through touch, and to be trained via touch.