RoboTera: Non-Contact Friction Sensing for Robotic Grasping via Wireless Sub-Terahertz Perception

摘要

Sensing friction coefficient is vital for various cyber-physical system applications, including robotic grasping. We present RoboTera, a novel system for the non-contact coefficient of friction (COF) estimation using sub-Terahertz (sub-THz) perception in robotics for the first time. While advanced tactile sensors can provide friction inputs, they require direct contact, which might not be suitable for various applications. Non-contact estimation of friction between the gripper and a target object requires extracting the minute surface perturbations which is unfortunately not supported by existing imaging modalities (such as camera and LiDAR). Our key insight is that sub-THz signals are best suited to infer such information as their sub-millimeter wavelength is comparable with surface perturbations. Hence, impinging sub-THz waves on everyday objects creates diffuse backscattering whose spectral profile hints at surface texture properties. Leveraging this, we use sub-THz wireless signals to extract surface roughness. By integrating sub-THz-estimated roughness inputs with conventional image-based material classification schemes, RoboTera provides a non-contact and precise COF inference framework. Further, we exploit COF inferences to identify stable grasp configurations and improve grasping performance. Our experiments demonstrate an average accuracy of over 92% in COF estimation. We implemented RoboTera on a robotic arm to assess its real-world grasping performance, achieving a 31.8% average improvement across objects with diverse COF profiles and shapes.