On Humanoid Robots Imitating Human Touch Gestures on the Smart Phone

We showcase an attack in which an autonomous humanoid robot is trained to execute touch gestures that match those of a target user. Different from past work which addressed a similar problem using a Lego robot, we harness the significant processing power and unique motoric capabilities of the autonomous humanoid robot to implement an attack that: (1) executes touch gestures with high precision, (2) is easily adapted to execute gestures on different touch screen devices, and (3) requires minimal human involvement. Relative to the traditional zero-effort impostor attacks, we show, based on a dataset of 26 users, that our attack significantly degrades the performance of touch-based authentication systems. In addition to the paper highlighting the threat that sophisticated adversaries pose to touch-based authentication systems, our robotic attack design provides a blueprint for much needed impostor testing mechanisms that simulate algorithmic (or sophisticated) adversaries against touch-based authentication systems.