Enabling Low-Cost Full Surface Tactile Skin for Human Robot Interaction

Realizing full coverage, low-maintenance, and low-cost tactile skin is a de facto design dream since the invention of robots. It ensures safety and enables collaborative work protocols for human robot interactions (HRI). The on-robot tactile capability is realized by deploying an array of external sensors or inferring from proprioceptive information that comes with the robot, such as motor torque. However, these methods may be cumbersome, introduce extra management cost, expensive, lack real-world robustness, or require special robot designs. In this letter, we present Sonic-Skin, a low-cost (< $2) and easy to deploy system that localizes the on-robot human touch and estimates the touch pressure without actually attaching sensors at potential touch locations. The system requires only a single pair of piezoelectric transducers (i.e. one transmitter and one receiver) attached on the target robot and turns the robot itself into a versatile sensor. We present a set of novel algorithms to progressively address the unique challenges posed by our system design. We put together an end-to-end SonicSkin system on a Jaco robot arm that runs in real-time, and conducted an extensive real-world study including 57019 actual evaluation datapoints under various challenging conditions from 12 human subjects. SonicSkin achieves less than 2 cm localization error for 96.4% of touches, with more than 96.7% cross-correlation similarity between the predicted touch pressure and the ground truth touch pressure.