A Vision-Based Tactile-Roller Grasper for Delicate Manipulation of Ultra-Soft Objects

Abstract

In-hand manipulation of ultra-soft objects is challenging due to their high deformability. Conventional vision-based tactile sensors (VBTSs) with reflective coatings often fail to detect contact signals from ultra-soft items like sponges, since the sensing material is stiffer than the object itself, resulting in insufficient deformation for reliable signal generation. This letter proposes a two-fingered grasper with steerable rollers integrated with a VBTS based on a light-conductive plate, achieving high sensitivity to weak contact forces. A self-calibration mechanism and a dynamic correspondence mechanism are developed to account for the rolling characteristics of the sensor. In addition, a Jacobian-based redundant control scheme is implemented to dynamically regulate clamping forces while maintaining manipulation trajectories. Experiments show that the system effectively decouples force regulation from manipulation for various sponges. Furthermore, through pure force-regulated manipulation, the grasper shows the capability to handle extremely compliant food items.