DESIGN OF A FOUR-FINGERED ADAPTIVE GRIPPER WITH RECONFIGURABILITY-BASED TACTILE FEATURES ARGUMENTATION

Abstract

For robotic grippers with limited degrees of freedom (DoF), grasping and manipulation of unstructured objects are provoking new challenges to both gripper design and control. In this thesis, we design a novel multi-fingered under-actuated gripper, integrated with whole-gripper tactile sensors. Two novel finger mechanisms are developed and analyzed. Further, combing the tactile perception and structural features, Graph Neural Networks (GNN) are applied to recognizing objects from a single grasp. We novelly augment the tactile feedback through tactile data rotation (TDR). The performance of the gripper system is experimentally evaluated. Results show that the proposed gripper can stably and compliantly grasp a wide range of daily objects with various sizes, shapes, and stiffness and manipulate them. Besides, TDR significantly improves the model's generalization ability for object recognition and increases the accuracy. Such a gripper system potentially benefits the service industry where stable grasping for various objects is needed.