Implementation and Evaluation of a Size-Adaptive Variable-Stiffness Soft Gripper Using Vision and Torque Monitoring

Abstract

Although conventional soft grippers are well-suited for handling delicate objects, they often exhibit limitations in agile automation, such as a reliance on bulky pneumatic systems or a lack of integrated sensing for adaptive control. To address these challenges, particularly in high-mix, low-volume (HMLV) environments, this paper presents a novel tendon-driven four-finger soft gripper with a variable stiffness mechanism. This work introduces an intelligent grasping paradigm that obviates the need for discrete fingertip sensors by leveraging a dual-modality approach, fusing data from an onboard camera and motor torque feedback via a rule-based dead-zone mechanism. This integrated sensing informs a data-driven kinematic model, based on second-order multivariate polynomial regression, to afford precise vision-guided control of the fingertips. Concurrently, this data fusion facilitates real-time incipient slip detection, which triggers an autonomous re-grasping sequence to maintain grasp stability while preventing potential damage to fragile objects. Experimental validation on a UR5e robotic arm demonstrated high success rates in pick-and-place tasks across six objects of diverse geometries and weights, along with robust execution of the re-grasping maneuver. These results highlight the system’s potential as a compact, intelligent, and cost-effective solution for versatile automation.